# How to use modelparams.dev [modelparams.dev](https://modelparams.dev) is an open, community-maintained catalog of LLM API parameters: the settings you send in a request, like `temperature` and `max_tokens`. Each entry shows the knobs you can turn — type, default, range, and the conditions that gate it. It does not track parameter counts. "Parameters" here never means trained weights, so nothing on this site answers how many parameters a model has. See https://modelparams.dev/model-parameters-vs-api-parameters for the difference. The same model accessed via an **API key** and via a **subscription** usually exposes a different set of parameters. We list both as separate entries so the data stays honest. A `wireId` containing `{scope}` needs one substitution before you send it: Bedrock reaches most newer models through a cross-region inference profile whose id is the model id behind a geography prefix, so `{scope}.anthropic.claude-sonnet-4-5-20250929-v1:0` becomes `eu.anthropic....` in Frankfurt, or `global.anthropic....` to let AWS route it. Valid scopes: us, eu, apac, jp, au, ca, sa, global. A wireId with no placeholder is already complete. Every entry names its request and SDK family in `apiSurface`; **one entry documents one wire format**. Where a host serves two, use the entry for the surface your code calls: `bedrock/*` currently declares `amazon-bedrock-converse` and documents Amazon Bedrock's `Converse` API (`inferenceConfig.maxTokens`, vendor extras under `additionalModelRequestFields`), not `InvokeModel` with native per-vendor bodies (`max_tokens`, top-level `thinking`) as the `AnthropicBedrock` client sends. If your code calls the surface an entry does not cover, its parameter names will not match. ## Catalog API The full catalog is static JSON, CORS-enabled, served from the edge: ```bash curl https://modelparams.dev/api/v1/models.json ``` Each entry is keyed by `provider/model` for API-key variants; subscription variants append `-subscription`. When you only need the parameter list for a model contract, use the providerless params endpoint. Subscription contracts are model slugs with `-subscription`: ```bash curl https://modelparams.dev/api/v1/models/openai/gpt-5.5.json curl https://modelparams.dev/api/v1/models/openai/gpt-5.5-subscription.json ``` ## Single model ```bash curl https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-7.json curl https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-7-subscription.json ``` ## Validate a request before you send it POST a model and a params object; the response reports unknown parameters, values outside their range, and combinations the provider rejects — plus a corrected payload under `safeParams` that is always safe to send as-is: ```bash curl -s https://modelparams.dev/api/v1/validate \ -H 'Content-Type: application/json' \ -d '{"model":"claude-3-opus-20240229","params":{"temperature":0.5,"top_p":0.9}}' ``` Each issue carries a `code`: `unknown_parameter` (no such knob on this model), `invalid_value` (out of range or not in the enum), or `not_applicable` (the knob exists but conflicts with another value in the same request, listed in `conflictsWith`). ## MCP server Agents can query the catalog over the Model Context Protocol instead of HTTP. The server is hosted, over Streamable HTTP — nothing to install: ```bash claude mcp add --transport http modelparams https://modelparams.dev/mcp codex mcp add modelparams --url https://modelparams.dev/mcp ``` Tools: `validate_model_params`, `get_model_params`, `list_models`, `list_provider_models`. ## JSON Schema Every entry validates against a JSON Schema you can use in your editor or pipeline: ```bash curl https://modelparams.dev/api/v1/schema.json ``` Add this header to any YAML you author for autocomplete in VS Code: ```yaml # yaml-language-server: $schema=https://modelparams.dev/api/v1/schema.json ``` ## Logos Provider logos are at `/assets/logos/{provider}.svg`. They use `currentColor`, so they inherit your text color: ```bash curl https://modelparams.dev/assets/logos/anthropic.svg ``` ## Contribute The data lives in YAML under `models/{provider}/{model}-{auth}.yaml` in the [GitHub repo](https://github.com/mnfst/modelparams.dev). Open a PR; CI validates against the schema and rebuilds. ## For agents - Machine-readable site overview: https://modelparams.dev/llms.txt - Full usage guide plus every parameter inline: https://modelparams.dev/llms-full.txt - When your browser supports it, this page registers in-browser **WebMCP** tools on `navigator.modelContext`: `search_models`, `get_model_parameters`, `list_providers`, `list_parameters`, and `get_usage_guide`. --- # Full catalog 412 models, grouped by provider. Each line reads: `path` (type, group) — description, then defaults, ranges, allowed values, and applicability conditions in brackets. ## Bedrock ### bedrock/claude-haiku-4-5 Bedrock Claude Haiku 4.5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/claude-haiku-4-5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/claude-opus-4-5-20251101 Bedrock Claude Opus 4.5 20251101 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/claude-opus-4-5-20251101.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/claude-opus-4-6 Bedrock Claude Opus 4.6 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/claude-opus-4-6.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/claude-sonnet-4-5 Bedrock Claude Sonnet 4.5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/claude-sonnet-4-5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/claude-sonnet-4-6 Bedrock Claude Sonnet 4.6 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/claude-sonnet-4-6.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/deepseek-r1 Bedrock DeepSeek R1 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/deepseek-r1.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/deepseek-v3.2 Bedrock DeepSeek v3.2 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/deepseek-v3.2.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/devstral-2-123b Bedrock Devstral 2 123B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/devstral-2-123b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gemma-3-12b-it Bedrock Gemma 3 12B IT on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gemma-3-12b-it.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gemma-3-27b-it Bedrock Gemma 3 27B IT on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gemma-3-27b-it.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gemma-3-4b-it Bedrock Gemma 3 4B IT on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gemma-3-4b-it.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/glm-4.7 Bedrock GLM-4.7 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/glm-4.7.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/glm-4.7-flash Bedrock GLM-4.7 Flash on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/glm-4.7-flash.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/glm-5 Bedrock GLM-5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/glm-5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gpt-oss-120b Bedrock GPT-OSS 120B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gpt-oss-120b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gpt-oss-20b Bedrock GPT-OSS 20B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gpt-oss-20b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gpt-oss-safeguard-120b Bedrock GPT-OSS Safeguard 120B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gpt-oss-safeguard-120b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/gpt-oss-safeguard-20b Bedrock GPT-OSS Safeguard 20B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/gpt-oss-safeguard-20b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/kimi-k2-thinking Bedrock Kimi K2 Thinking on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/kimi-k2-thinking.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/kimi-k2.5 Bedrock Kimi K2.5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/kimi-k2.5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/llama3-1-70b-instruct Bedrock Llama3 1 70B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/llama3-1-70b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/llama3-1-8b-instruct Bedrock Llama3 1 8B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/llama3-1-8b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/llama3-3-70b-instruct Bedrock Llama3 3 70B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/llama3-3-70b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/llama3-70b-instruct Bedrock Llama3 70B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/llama3-70b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/llama3-8b-instruct Bedrock Llama3 8B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/llama3-8b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/magistral-small-2509 Bedrock Magistral Small 2509 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/magistral-small-2509.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/minimax-m2 Bedrock MiniMax M2 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/minimax-m2.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/minimax-m2.1 Bedrock MiniMax M2.1 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/minimax-m2.1.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/minimax-m2.5 Bedrock MiniMax M2.5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/minimax-m2.5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/ministral-3-14b-instruct Bedrock Ministral 3 14B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/ministral-3-14b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] - `additionalModelRequestFields.thinking.type` (enum, Reasoning) — Controls Anthropic extended thinking on Bedrock. Bedrock rejects the adaptive mode the first-party Anthropic API accepts, so only disabled and enabled are listed here. [default: "disabled"] [values: "disabled", "enabled"] - `additionalModelRequestFields.thinking.budget_tokens` (integer, Reasoning) — Maximum token budget the model may use for extended thinking before producing the final answer. Must be less than inferenceConfig.maxTokens. [range: min 1024] [only when additionalModelRequestFields.thinking.type = "enabled"] ### bedrock/ministral-3-3b-instruct Bedrock Ministral 3 3B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/ministral-3-3b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/ministral-3-8b-instruct Bedrock Ministral 3 8B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/ministral-3-8b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/mistral-7b-instruct Bedrock Mistral 7B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/mistral-7b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] ### bedrock/mistral-large-2402 Bedrock Mistral Large 2402 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/mistral-large-2402.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] ### bedrock/mistral-large-3-675b-instruct Bedrock Mistral Large 3 675B Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/mistral-large-3-675b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/mistral-small-2402 Bedrock Mistral Small 2402 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/mistral-small-2402.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] ### bedrock/mixtral-8x7b-instruct Bedrock Mixtral 8x7b Instruct on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/mixtral-8x7b-instruct.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Converse carries it through additionalModelRequestFields; it is not part of inferenceConfig. [range: min 0] ### bedrock/nemotron-nano-12b Bedrock Nemotron Nano 12B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nemotron-nano-12b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/nemotron-nano-3-30b Bedrock Nemotron Nano 3 30B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nemotron-nano-3-30b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/nemotron-nano-9b Bedrock Nemotron Nano 9B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nemotron-nano-9b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/nemotron-super-3-120b Bedrock Nemotron Super 3 120B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nemotron-super-3-120b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/nova-2-lite Bedrock Nova 2 Lite on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nova-2-lite.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.inferenceConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Nova nests it under an inferenceConfig object inside additionalModelRequestFields. [range: min 1] ### bedrock/nova-lite Bedrock Nova Lite on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nova-lite.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.inferenceConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Nova nests it under an inferenceConfig object inside additionalModelRequestFields. [range: min 1] ### bedrock/nova-micro Bedrock Nova Micro on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nova-micro.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.inferenceConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Nova nests it under an inferenceConfig object inside additionalModelRequestFields. [range: min 1] ### bedrock/nova-pro Bedrock Nova Pro on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/nova-pro.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] - `additionalModelRequestFields.inferenceConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. Nova nests it under an inferenceConfig object inside additionalModelRequestFields. [range: min 1] ### bedrock/palmyra-vision-7b Bedrock Palmyra Vision 7B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/palmyra-vision-7b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/palmyra-x4 Bedrock Palmyra X4 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/palmyra-x4.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/palmyra-x5 Bedrock Palmyra X5 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/palmyra-x5.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/pixtral-large-2502 Bedrock Pixtral Large 2502 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/pixtral-large-2502.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/qwen3-32b Bedrock Qwen3 32B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/qwen3-32b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/qwen3-coder-30b-a3b Bedrock Qwen3 Coder 30B A3B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/qwen3-coder-30b-a3b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/qwen3-coder-next Bedrock Qwen3 Coder Next on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/qwen3-coder-next.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/qwen3-next-80b-a3b Bedrock Qwen3 Next 80B A3B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/qwen3-next-80b-a3b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/qwen3-vl-235b-a22b Bedrock Qwen3 Vl 235B A22B on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/qwen3-vl-235b-a22b.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/voxtral-mini-3b-2507 Bedrock Voxtral Mini 3B 2507 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/voxtral-mini-3b-2507.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ### bedrock/voxtral-small-24b-2507 Bedrock Voxtral Small 24B 2507 on Converse API · JSON: https://modelparams.dev/api/v1/models/bedrock/voxtral-small-24b-2507.json - `inferenceConfig.maxTokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `inferenceConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `inferenceConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Bedrock rejects values above 1 for this model. [range: min 0, max 1, step 0.1] - `inferenceConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [range: min 0, max 1, step 0.01] ## OpenAI ### openai/chatgpt-4o-latest OpenAI ChatGPT 4o Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/chatgpt-4o-latest.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-3.5-turbo OpenAI GPT-3.5 Turbo on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-3.5-turbo.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-3.5-turbo-16k OpenAI GPT-3.5 Turbo 16K on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-3.5-turbo-16k.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4 OpenAI GPT-4 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high"] - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### openai/gpt-4-0613 OpenAI GPT-4 0613 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4-0613.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4-turbo OpenAI GPT-4 Turbo on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4-turbo.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4-turbo-2024-04-09 OpenAI GPT-4 Turbo 2024-04-09 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4-turbo-2024-04-09.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4.1 OpenAI GPT-4.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4.1.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4.1-mini OpenAI GPT-4.1 Mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4.1-mini.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4.1-nano OpenAI GPT-4.1 Nano on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4.1-nano.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4o OpenAI GPT-4o on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4o.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4o-2024-11-20 OpenAI GPT-4o 2024-11-20 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4o-2024-11-20.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-4o-mini OpenAI GPT-4o mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-4o-mini.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] ### openai/gpt-5 OpenAI GPT-5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] ### openai/gpt-5-chat-latest OpenAI GPT-5 Chat Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5-chat-latest.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] ### openai/gpt-5-mini OpenAI GPT-5 Mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5-mini.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] ### openai/gpt-5-nano OpenAI GPT-5 Nano on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5-nano.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] ### openai/gpt-5.1 OpenAI GPT-5.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.1.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "none"] [values: "none", "low", "medium", "high"] ### openai/gpt-5.1-codex-max-subscription OpenAI GPT-5.1 Codex Max (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.1-codex-max-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.1-codex-subscription OpenAI GPT-5.1 Codex (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.1-codex-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.2 OpenAI GPT-5.2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.2.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.2-codex-subscription OpenAI GPT-5.2 Codex (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.2-codex-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.2-subscription OpenAI GPT-5.2 (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.2-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.3-codex OpenAI GPT-5.3 Codex on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.3-codex.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ### openai/gpt-5.3-codex-spark-subscription OpenAI GPT-5.3 Codex Spark (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.3-codex-spark-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.3-codex-subscription OpenAI GPT-5.3 Codex (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.3-codex-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.4 OpenAI GPT-5.4 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.4-mini OpenAI GPT-5.4 Mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-mini.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.4-mini-subscription OpenAI GPT-5.4 Mini (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-mini-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.4-nano OpenAI GPT-5.4 Nano on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-nano.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.4-pro OpenAI GPT-5.4 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-pro.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "medium", "high", "xhigh"] ### openai/gpt-5.4-pro-subscription OpenAI GPT-5.4 Pro (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-pro-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.4-subscription OpenAI GPT-5.4 (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.4-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.5 OpenAI GPT-5.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.5.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.5-pro OpenAI GPT-5.5 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.5-pro.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "medium", "high", "xhigh"] ### openai/gpt-5.5-pro-subscription OpenAI GPT-5.5 Pro (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.5-pro-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.5-subscription OpenAI GPT-5.5 (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.5-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high", "xhigh"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed", "none"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.6 OpenAI GPT-5.6 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "none"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.6-luna OpenAI GPT-5.6 Luna on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-luna.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "none"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.6-luna-subscription OpenAI GPT-5.6 Luna (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-luna-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh", "max"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.6-sol OpenAI GPT-5.6 Sol on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-sol.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "none"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.6-sol-subscription OpenAI GPT-5.6 Sol (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-sol-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh", "max"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-5.6-terra OpenAI GPT-5.6 Terra on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-terra.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "none"] [values: "none", "low", "medium", "high", "xhigh"] ### openai/gpt-5.6-terra-subscription OpenAI GPT-5.6 Terra (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/openai/gpt-5.6-terra-subscription.json - `reasoning.effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "none", "low", "medium", "high", "xhigh", "max"] - `reasoning.summary` (enum, Reasoning) — Controls the level of reasoning summary returned with the response. [default: "auto"] [values: "auto", "concise", "detailed"] - `text.verbosity` (enum, Output) — Controls how concise or detailed the model's final text response should be. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-6-astra OpenAI GPT-6 Astra on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-6-astra.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high"] - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### openai/gpt-oss-120b OpenAI GPT-OSS 120B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-oss-120b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high"] - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### openai/gpt-oss-20b OpenAI GPT-OSS 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-oss-20b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high"] - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### openai/gpt-oss-safeguard-120b OpenAI GPT-OSS Safeguard 120B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-oss-safeguard-120b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model performs when interpreting the provided safety policy before returning a classification. [default: "medium"] [values: "low", "medium", "high"] ### openai/gpt-oss-safeguard-20b OpenAI GPT-OSS Safeguard 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/gpt-oss-safeguard-20b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. OpenAI recommends sampling at 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. OpenAI recommends sampling at 1.0. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model performs when interpreting the provided safety policy before returning a classification. [default: "medium"] [values: "low", "medium", "high"] ### openai/o1 OpenAI o1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o1.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ### openai/o1-mini OpenAI o1-mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o1-mini.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] ### openai/o1-preview OpenAI o1-preview on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o1-preview.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "minimal", "low", "medium", "high"] ### openai/o3 OpenAI o3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o3.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ### openai/o3-mini OpenAI o3-mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o3-mini.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ### openai/o3-pro OpenAI o3-pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o3-pro.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ### openai/o4-mini OpenAI o4-mini on Chat Completions · JSON: https://modelparams.dev/api/v1/models/openai/o4-mini.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 16] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [default: "medium"] [values: "low", "medium", "high", "xhigh"] ## Alibaba ### alibaba/deepseek-v3.2 Alibaba DeepSeek v3.2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/deepseek-v3.2.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/deepseek-v4-flash Alibaba DeepSeek v4 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/deepseek-v4-flash.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/deepseek-v4-flash-0731 Alibaba DeepSeek v4 Flash 0731 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/deepseek-v4-flash-0731.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/deepseek-v4-pro Alibaba DeepSeek v4 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/deepseek-v4-pro.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/deepseek-v4-pro-0813 Alibaba DeepSeek v4 Pro 0813 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/deepseek-v4-pro-0813.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/glm-5.1 Alibaba GLM-5.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/glm-5.1.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/glm-5.2 Alibaba GLM-5.2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/glm-5.2.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/kimi-k2.7-code Alibaba Kimi K2.7 Code on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/kimi-k2.7-code.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/kimi-k3 Alibaba Kimi K3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/kimi-k3.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen-flash Alibaba Qwen Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen-flash.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen-max Alibaba Qwen Max on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen-max.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen-plus Alibaba Qwen Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen-plus.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen-turbo Alibaba Qwen Turbo on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen-turbo.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3-235b-a22b-thinking-2507 Alibaba Qwen3 235B A22B Thinking 2507 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-235b-a22b-thinking-2507.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-30b-a3b-instruct-2507 Alibaba Qwen3 30B A3B Instruct 2507 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-30b-a3b-instruct-2507.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-30b-a3b-thinking-2507 Alibaba Qwen3 30B A3B Thinking 2507 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-30b-a3b-thinking-2507.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-coder-flash Alibaba Qwen3 Coder Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-coder-flash.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] ### alibaba/qwen3-coder-next Alibaba Qwen3 Coder Next on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-coder-next.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] ### alibaba/qwen3-coder-plus Alibaba Qwen3 Coder Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-coder-plus.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] ### alibaba/qwen3-max Alibaba Qwen3 Max on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-max.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-next-80b-a3b-instruct Alibaba Qwen3 Next 80B A3B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-next-80b-a3b-instruct.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-next-80b-a3b-thinking Alibaba Qwen3 Next 80B A3B Thinking on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-next-80b-a3b-thinking.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-vl-235b-a22b-instruct Alibaba Qwen3 Vl 235B A22B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-vl-235b-a22b-instruct.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3-vl-235b-a22b-thinking Alibaba Qwen3 Vl 235B A22B Thinking on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3-vl-235b-a22b-thinking.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: false] ### alibaba/qwen3.5 Alibaba Qwen3.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.5-122b-a10b Alibaba Qwen3.5 122B A10B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5-122b-a10b.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.5-27b Alibaba Qwen3.5 27B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5-27b.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.5-35b-a3b Alibaba Qwen3.5 35B A3B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5-35b-a3b.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.5-397b-a17b Alibaba Qwen3.5 397B A17B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5-397b-a17b.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.5-flash Alibaba Qwen3.5 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.5-flash.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] - `extra_body.chat_template_kwargs.enable_thinking` (boolean, Reasoning) — Controls Qwen3 thinking mode when using OpenAI-compatible clients that pass provider-specific extra body fields. [default: true] ### alibaba/qwen3.6-27b Alibaba Qwen3.6 27B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.6-27b.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.6-35b-a3b Alibaba Qwen3.6 35B A3B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.6-35b-a3b.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.6-flash Alibaba Qwen3.6 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.6-flash.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.6-max-preview Alibaba Qwen3.6 Max Preview on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.6-max-preview.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.6-plus Alibaba Qwen3.6 Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.6-plus.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.7-flash Alibaba Qwen3.7 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.7-flash.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.7-max Alibaba Qwen3.7 Max on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.7-max.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.7-plus Alibaba Qwen3.7 Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.7-plus.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.8-2.4t-a95b Alibaba Qwen3.8 2.4t A95B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.8-2.4t-a95b.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.8-27b Alibaba Qwen3.8 27B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.8-27b.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.8-flash Alibaba Qwen3.8 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.8-flash.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.8-max Alibaba Qwen3.8 Max on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.8-max.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwen3.8-max-0902 Alibaba Qwen3.8 Max 0902 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwen3.8-max-0902.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, including both reasoning and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. Values above 100 disable top-k sampling. [default: 20] [range: min 0] - `extra_body.enable_thinking` (boolean, Reasoning) — Toggles the model's hybrid thinking mode, sent as a provider-specific extra body field on OpenAI-compatible clients. [default: true] - `extra_body.thinking_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before it starts the final answer; defaults to the model's maximum reasoning length. [range: min 1] [only when extra_body.enable_thinking = true] ### alibaba/qwq-plus Alibaba QwQ Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/alibaba/qwq-plus.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `extra_body.top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 20] [range: min 1] ## Anthropic ### anthropic/claude-3-5-haiku-20241022 Anthropic Claude 3.5 Haiku 20241022 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-5-haiku-20241022.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-3-5-haiku-latest Anthropic Claude 3.5 Haiku Latest on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-5-haiku-latest.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-3-5-sonnet-20241022 Anthropic Claude 3.5 Sonnet 20241022 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-5-sonnet-20241022.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-3-5-sonnet-latest Anthropic Claude 3.5 Sonnet Latest on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-5-sonnet-latest.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-3-7-sonnet-20250219 Anthropic Claude 3.7 Sonnet 20250219 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-7-sonnet-20250219.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-3-7-sonnet-latest Anthropic Claude 3.7 Sonnet Latest on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-7-sonnet-latest.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-3-opus-20240229 Anthropic Claude 3 Opus 20240229 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-opus-20240229.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-3-opus-latest Anthropic Claude 3 Opus Latest on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-3-opus-latest.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] ### anthropic/claude-fable-5 Anthropic Claude Fable 5 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-fable-5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Only adaptive thinking is supported; omit the parameter entirely to run without thinking (an explicit disabled value is rejected). [values: "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-fable-5-1 Anthropic Claude Fable 5.1 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-fable-5-1.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Only adaptive thinking is supported; omit the parameter entirely to run without thinking (an explicit disabled value is rejected). [values: "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-fable-5-subscription Anthropic Claude Fable 5 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-fable-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Only adaptive thinking is supported; omit the parameter entirely to run without thinking (an explicit disabled value is rejected). [values: "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-haiku-4 Anthropic Claude Haiku 4 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-haiku-4-5 Anthropic Claude Haiku 4.5 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4-5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-haiku-4-5-20251001 Anthropic Claude Haiku 4.5 20251001 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4-5-20251001.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-haiku-4-5-20251001-subscription Anthropic Claude Haiku 4.5 20251001 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4-5-20251001-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-haiku-4-5-subscription Anthropic Claude Haiku 4.5 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-haiku-4-subscription Anthropic Claude Haiku 4 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-haiku-4-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-opus-4-1-20250805 Anthropic Claude Opus 4.1 20250805 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-1-20250805.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-opus-4-1-20250805-subscription Anthropic Claude Opus 4.1 20250805 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-1-20250805-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-opus-4-20250514 Anthropic Claude Opus 4 20250514 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-20250514.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-opus-4-20250514-subscription Anthropic Claude Opus 4 20250514 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-20250514-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-opus-4-5-20251101 Anthropic Claude Opus 4.5 20251101 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-5-20251101.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high"] ### anthropic/claude-opus-4-5-20251101-subscription Anthropic Claude Opus 4.5 20251101 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-5-20251101-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high"] ### anthropic/claude-opus-4-6 Anthropic Claude Opus 4.6 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-6.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"enabled", "adaptive"}; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"enabled", "adaptive"}; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"enabled", "adaptive"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type ∈ {"adaptive", "enabled"}] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "max"] ### anthropic/claude-opus-4-6-subscription Anthropic Claude Opus 4.6 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-6-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"enabled", "adaptive"}; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"enabled", "adaptive"}; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"enabled", "adaptive"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type ∈ {"adaptive", "enabled"}] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "max"] ### anthropic/claude-opus-4-7 Anthropic Claude Opus 4.7 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-7.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-opus-4-7-subscription Anthropic Claude Opus 4.7 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-7-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-opus-4-8 Anthropic Claude Opus 4.8 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-8.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-opus-4-8-subscription Anthropic Claude Opus 4.8 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-8-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-opus-4-subscription Anthropic Claude Opus 4 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-4-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-opus-5 Anthropic Claude Opus 5 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "adaptive"] [values: "disabled", "adaptive"] [except when output_config.effort ∈ {"xhigh", "max"}] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-opus-5-subscription Anthropic Claude Opus 5 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-opus-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "adaptive"] [values: "disabled", "adaptive"] [except when output_config.effort ∈ {"xhigh", "max"}] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "omitted"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-sonnet-4-20250514 Anthropic Claude Sonnet 4 20250514 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-20250514.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-20250514-subscription Anthropic Claude Sonnet 4 20250514 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-20250514-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-5 Anthropic Claude Sonnet 4.5 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-5-20250929 Anthropic Claude Sonnet 4.5 20250929 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-5-20250929.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-5-20250929-subscription Anthropic Claude Sonnet 4.5 20250929 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-5-20250929-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type = "enabled"; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-5-subscription Anthropic Claude Sonnet 4.5 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-4-6 Anthropic Claude Sonnet 4.6 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-6.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"enabled", "adaptive"}; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"enabled", "adaptive"}; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"enabled", "adaptive"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type ∈ {"adaptive", "enabled"}] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "max"] ### anthropic/claude-sonnet-4-6-subscription Anthropic Claude Sonnet 4.6 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-6-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"enabled", "adaptive"}; top_p ≠ null] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"enabled", "adaptive"}; temperature ≠ null] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"enabled", "adaptive"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type ∈ {"adaptive", "enabled"}] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "max"] ### anthropic/claude-sonnet-4-subscription Anthropic Claude Sonnet 4 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-4-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when thinking.type ∈ {"adaptive", "enabled"}] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens whose cumulative probability reaches this value. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type ∈ {"adaptive", "enabled"}; temperature ≠ 1] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 0] [range: min 0] [except when thinking.type ∈ {"adaptive", "enabled"}] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive", "enabled"] - `thinking.budget_tokens` (integer, Reasoning) — Maximum token budget Anthropic may use for extended thinking before producing the final answer. [default: 4096] [range: min 1024] [only when thinking.type = "enabled"] ### anthropic/claude-sonnet-5 Anthropic Claude Sonnet 5 on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-5.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ### anthropic/claude-sonnet-5-subscription Anthropic Claude Sonnet 5 (Subscription) on Messages API · JSON: https://modelparams.dev/api/v1/models/anthropic/claude-sonnet-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `thinking.type` (enum, Reasoning) — Controls the Anthropic thinking mode values supported by this model. [default: "disabled"] [values: "disabled", "adaptive"] - `thinking.display` (enum, Reasoning) — Controls whether Anthropic returns summarized or omitted thinking content. [default: "summarized"] [values: "summarized", "omitted"] [only when thinking.type = "adaptive"] - `output_config.effort` (enum, Reasoning) — Controls Anthropic response thoroughness and token spend. [default: "high"] [values: "low", "medium", "high", "xhigh", "max"] ## Google ### google/gemini-2.5-flash Google Gemini 2.5 Flash on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; 0 disables thinking and -1 uses dynamic thinking. [default: -1] [range: min -1, max 24576] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-2.5-flash-lite Google Gemini 2.5 Flash Lite on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; -1 uses dynamic thinking, 0 disables thinking, and fixed budgets start at 512 tokens. [default: 0] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-2.5-flash-lite-subscription Google Gemini 2.5 Flash Lite (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-flash-lite-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; -1 uses dynamic thinking, 0 disables thinking, and fixed budgets start at 512 tokens. [default: 0] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-2.5-flash-subscription Google Gemini 2.5 Flash (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-flash-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; 0 disables thinking and -1 uses dynamic thinking. [default: -1] [range: min -1, max 24576] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-2.5-pro Google Gemini 2.5 Pro on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-pro.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Maximum number of thinking tokens Gemini should use before producing the final answer. [range: min 128, max 32768] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-2.5-pro-subscription Google Gemini 2.5 Pro (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-2.5-pro-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Maximum number of thinking tokens Gemini should use before producing the final answer. [range: min 128, max 32768] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3-flash-preview Google Gemini 3 Flash Preview on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3-flash-preview.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3-flash-preview-subscription Google Gemini 3 Flash Preview (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3-flash-preview-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3 Flash reasoning effort. [default: "high"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-flash-lite Google Gemini 3.1 Flash Lite on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-flash-lite-preview Google Gemini 3.1 Flash Lite Preview on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-flash-lite-preview.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-flash-lite-preview-subscription Google Gemini 3.1 Flash Lite Preview (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-flash-lite-preview-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "high"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-flash-lite-subscription Google Gemini 3.1 Flash Lite (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-flash-lite-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-pro-preview Google Gemini 3.1 Pro Preview on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-pro-preview.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-pro-preview-customtools Google Gemini 3.1 Pro Preview Customtools on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-pro-preview-customtools.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.1-pro-preview-subscription Google Gemini 3.1 Pro Preview (Subscription) on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.1-pro-preview-subscription.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3 Pro reasoning effort. [default: "high"] [values: "low", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.5-flash Google Gemini 3.5 Flash on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.5-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.5 Flash reasoning effort. [default: "medium"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.5-flash-lite Google Gemini 3.5 Flash Lite on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.5-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.5 Flash reasoning effort. [default: "medium"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.6-flash Google Gemini 3.6 Flash on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.6-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.7-flash Google Gemini 3.7 Flash on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.7-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-3.8-flash Google Gemini 3.8 Flash on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-3.8-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemini-flash-latest Google Gemini Flash Latest on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemini-flash-latest.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; -1 uses dynamic thinking, 0 disables thinking, and fixed budgets start at 512 tokens. [default: 0] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemma-3-12b-it Google Gemma 3 12B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3-12b-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 128K context window with the input. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-3-1b-it Google Gemma 3 1B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3-1b-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 32K context window with the input. [range: min 1, max 32768] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-3-27b-it Google Gemma 3 27B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3-27b-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 128K context window with the input. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-3-4b-it Google Gemma 3 4B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3-4b-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 128K context window with the input. [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-3n-E2B-it Google Gemma 3n E2B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3n-E2B-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 32K context window with the input. [range: min 1, max 32768] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-3n-E4B-it Google Gemma 3n E4B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-3n-E4B-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate; output shares the 32K context window with the input. [range: min 1, max 32768] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-4-12B-it Google Gemma 4 12B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-4-12B-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Google's standardized Gemma 4 sampling uses 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-4-26b-a4b-it Google Gemma 4 26B A4B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-4-26b-a4b-it.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; -1 uses dynamic thinking, 0 disables thinking, and fixed budgets start at 512 tokens. [default: 0] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Toggles Gemma 4 reasoning; high enables thinking and minimal disables it. [values: "minimal", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemma-4-31b-it Google Gemma 4 31B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-4-31b-it.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 0] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Number of thinking tokens Gemini should use; -1 uses dynamic thinking, 0 disables thinking, and fixed budgets start at 512 tokens. [default: 0] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Toggles Gemma 4 reasoning; high enables thinking and minimal disables it. [values: "minimal", "high"] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### google/gemma-4-E2B-it Google Gemma 4 E2B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-4-E2B-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Google's standardized Gemma 4 sampling uses 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ### google/gemma-4-E4B-it Google Gemma 4 E4B IT on Generate content · JSON: https://modelparams.dev/api/v1/models/google/gemma-4-E4B-it.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Google's standardized Gemma 4 sampling uses 1.0. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1] - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. [default: 64] [range: min 0] ## NVIDIA ### nvidia/deepseek-v4-flash-0731 NVIDIA DeepSeek v4 Flash 0731 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/deepseek-v4-flash-0731.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/deepseek-v4-pro-0813 NVIDIA DeepSeek v4 Pro 0813 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/deepseek-v4-pro-0813.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/gemma-4-31b-it NVIDIA Gemma 4 31B IT on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/gemma-4-31b-it.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/gliner-pii NVIDIA GLiNER PII on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/gliner-pii.json - `threshold` (number, Sampling) — Confidence threshold for entity detection. Lower values detect more entities but may include false positives. [default: 0.5] [range: min 0, max 1] - `chunk_length` (integer, Metadata) — Context window size for processing. Longer texts are automatically split into chunks with overlap for complete coverage. Must be greater than overlap. [default: 384] [range: min 1, max 2048] - `overlap` (integer, Metadata) — Token overlap between chunks to prevent entity clipping. Must be less than chunk_length. [default: 128] [range: min 0, max 512] - `flat_ner` (boolean, Metadata) — When true, prevents overlapping entity spans. When false, may return nested entities such as both a full name and its constituent first name. [default: false] ### nvidia/gpt-oss-120b NVIDIA GPT-OSS 120B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/gpt-oss-120b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/gpt-oss-20b NVIDIA GPT-OSS 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/gpt-oss-20b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/kimi-k3 NVIDIA Kimi K3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/kimi-k3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/laguna-xs-2.1 NVIDIA Laguna Xs 2.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/laguna-xs-2.1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/llama-3.1-nemoguard-8b-topic-control NVIDIA Llama 3.1 NemoGuard 8B Topic Control on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.1-nemoguard-8b-topic-control.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 1024] [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.5] [range: min 0, max 2] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] ### nvidia/llama-3.1-nemotron-nano-8b-v1 NVIDIA Llama 3.1 Nemotron Nano 8B v1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.1-nemotron-nano-8b-v1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/llama-3.1-nemotron-safety-guard-8b-v3 NVIDIA Llama 3.1 Nemotron Safety Guard 8B v3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.1-nemotron-safety-guard-8b-v3.json - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0] [range: min 0, max 1] ### nvidia/llama-3.1-nemotron-ultra-253b-v1 NVIDIA Llama 3.1 Nemotron Ultra 253B v1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.1-nemotron-ultra-253b-v1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/llama-3.3-nemotron-super-49b-v1 NVIDIA Llama 3.3 Nemotron Super 49B v1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.3-nemotron-super-49b-v1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/llama-3.3-nemotron-super-49b-v1.5 NVIDIA Llama 3.3 Nemotron Super 49B v1.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/llama-3.3-nemotron-super-49b-v1.5.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 65536] [range: min 1, max 65536] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/minimax-m3 NVIDIA MiniMax M3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/minimax-m3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/muse-glimmer-30b NVIDIA Muse Glimmer 30B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/muse-glimmer-30b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 4096] [range: min 1, max 16384] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.6] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Changing the seed produces a different response with similar characteristics. Fix the seed to reproduce results. [default: 0] [range: min 0, max 18446744073709552000] ### nvidia/nemoguard-jailbreak-detect NVIDIA NemoGuard Jailbreak Detect on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemoguard-jailbreak-detect.json _No parameters documented yet._ ### nvidia/nemotron-3-nano-30b-a3b NVIDIA Nemotron 3 Nano 30B A3B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-3-nano-30b-a3b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 16384] [range: min 1, max 32768] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Repeated requests with the same seed and parameters should return the same result. [range: min 0, max 18446744073709552000] ### nvidia/nemotron-3-super-120b-a12b NVIDIA Nemotron 3 Super 120B A12B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-3-super-120b-a12b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 16384] [range: min 1, max 32768] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Repeated requests with the same seed and parameters should return the same result. [range: min 0, max 18446744073709552000] - `reasoning_effort` (enum, Reasoning) — Controls the reasoning mode. 'none' disables reasoning tokens, 'low' enables low-effort reasoning, and 'high' enables full reasoning. [default: "high"] [values: "none", "low", "high"] - `reasoning_budget` (integer, Reasoning) — Maximum number of tokens the model may use for internal reasoning before being forced to end the reasoning trace. Use -1 to disable budget enforcement. [default: 16384] [range: min -1, max 32768] ### nvidia/nemotron-3-ultra-550b-a55b NVIDIA Nemotron 3 Ultra 550B A55B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-3-ultra-550b-a55b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 16384] [range: min 1, max 32768] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Repeated requests with the same seed and parameters should return the same result. [range: min 0, max 18446744073709552000] - `reasoning_effort` (enum, Reasoning) — Controls the reasoning mode. 'none' disables reasoning tokens, 'medium' enables efficient reasoning, and 'high' enables full reasoning. [default: "high"] [values: "none", "medium", "high"] - `reasoning_budget` (integer, Reasoning) — Maximum number of tokens the model may use for internal reasoning before being forced to end the reasoning trace. Use -1 to disable budget enforcement. [default: 16384] [range: min -1, max 32768] ### nvidia/nemotron-3-ultra-subscription NVIDIA Nemotron 3 Ultra (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-3-ultra-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 16384] [range: min 1, max 32768] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.95] [range: max 1] - `reasoning_effort` (enum, Reasoning) — Controls the reasoning mode. 'none' disables reasoning tokens, 'medium' enables efficient reasoning, and 'high' enables full reasoning. [default: "high"] [values: "none", "medium", "high"] - `reasoning_budget` (integer, Reasoning) — Maximum number of tokens the model may use for internal reasoning before being forced to end the reasoning trace. Use -1 to disable budget enforcement. [default: 16384] [range: min -1, max 32768] ### nvidia/nemotron-content-safety-reasoning-4b NVIDIA Nemotron Content Safety Reasoning 4B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-content-safety-reasoning-4b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 16384] [range: min 1, max 32768] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `seed` (integer, Sampling) — Best-effort deterministic sampling seed. Repeated requests with the same seed and parameters should return the same result. [range: min 0, max 18446744073709552000] ### nvidia/nemotron-mini-4b-instruct NVIDIA Nemotron Mini 4B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/nemotron-mini-4b-instruct.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 1024] [range: min 1, max 4096] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.2] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.7] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] ### nvidia/riva-translate-4b-instruct-v1.1 NVIDIA Riva Translate 4B Instruct v1.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/riva-translate-4b-instruct-v1.1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 512] [range: min 1, max 4096] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. The returned text will not contain the stop sequence. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 0.9] [range: max 1] - `frequency_penalty` (number, Sampling) — Penalizes new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim. [default: 0] [range: min -2, max 2] - `presence_penalty` (number, Sampling) — Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics. [default: 0] [range: min -2, max 2] ### nvidia/usdcode-llama-3.1-70b-instruct NVIDIA USDCode Llama 3.1 70B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/nvidia/usdcode-llama-3.1-70b-instruct.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate. Generation stops when this limit is reached. [default: 1024] [range: min 1, max 2048] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Not recommended to modify both temperature and top_p in the same call. [default: 0.1] [range: min 0, max 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. Not recommended to modify both temperature and top_p in the same call. [default: 1] [range: max 1] - `expert_type` (enum, Metadata) — The type of expert to use. 'knowledge' answers with USD knowledge, 'code' responds with vanilla OpenUSD code, 'helperfunction' uses high-level helper functions, and 'auto' lets the LLM determine which expert to use. [default: "auto"] [values: "auto", "code", "knowledge", "helperfunction"] ## Z.ai ### z-ai/glm-4.5 Z.ai GLM-4.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-air Z.ai GLM-4.5-Air on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-air.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-air-subscription Z.ai GLM-4.5-Air (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-air-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-airx Z.ai GLM-4.5-AirX on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-airx.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-flash Z.ai GLM-4.5-Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-flash.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-subscription Z.ai GLM-4.5 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.5-x Z.ai GLM-4.5-X on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.5-x.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 98304] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.6] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.6 Z.ai GLM-4.6 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.6.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.6-subscription Z.ai GLM-4.6 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.6-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.7 Z.ai GLM-4.7 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.7.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.7-flash Z.ai GLM-4.7-Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.7-flash.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.7-flashx Z.ai GLM-4.7-FlashX on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.7-flashx.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-4.7-subscription Z.ai GLM-4.7 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-4.7-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5 Z.ai GLM-5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5-subscription Z.ai GLM-5 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5-turbo Z.ai GLM-5-Turbo on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5-turbo.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5-turbo-subscription Z.ai GLM-5-Turbo (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5-turbo-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.1 Z.ai GLM-5.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.1.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.1-subscription Z.ai GLM-5.1 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.1-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.2 Z.ai GLM-5.2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.2.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning effort GLM-5.2 spends when thinking is enabled. [default: "max"] [values: "none", "minimal", "low", "medium", "high", "xhigh", "max"] [only when thinking.type = "enabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.2-subscription Z.ai GLM-5.2 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.2-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — Toggles the model's extended reasoning before it produces the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning effort GLM-5.2 spends when thinking is enabled. [default: "max"] [values: "none", "minimal", "low", "medium", "high", "xhigh", "max"] [only when thinking.type = "enabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.3 Z.ai GLM-5.3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — GLM-5.3 always engages in extended reasoning; thinking cannot be disabled. [default: "enabled"] [values: "enabled"] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning effort GLM-5.3 spends on its always-on thinking. [default: "max"] [values: "low", "high", "max"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### z-ai/glm-5.3-flash Z.ai GLM-5.3 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/z-ai/glm-5.3-flash.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [default: 65536] [range: min 1, max 131072] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 1, step 0.1] [except when do_sample = false] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] [except when do_sample = false] - `do_sample` (boolean, Sampling) — When false, the model uses greedy decoding and ignores temperature and top_p. [default: true] - `thinking.type` (enum, Reasoning) — GLM-5.3 always engages in extended reasoning; thinking cannot be disabled. [default: "enabled"] [values: "enabled"] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning effort GLM-5.3 spends on its always-on thinking. [default: "max"] [values: "low", "high", "max"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ## Mistral ### mistral/codestral-2508 Mistral Codestral 2508 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/codestral-2508.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/codestral-latest Mistral Codestral Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/codestral-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/devstral-2512 Mistral Devstral 2512 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/devstral-2512.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/devstral-latest Mistral Devstral Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/devstral-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/glm-5-2 Mistral GLM-5.2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/glm-5-2.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `prompt_mode` (enum, Reasoning) — Enables Mistral's reasoning system prompt; leave unset to disable the default reasoning behavior. [values: "reasoning"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/magistral-medium-latest Mistral Magistral Medium Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/magistral-medium-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `prompt_mode` (enum, Reasoning) — Enables Mistral's reasoning system prompt; leave unset to disable the default reasoning behavior. [values: "reasoning"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/magistral-small-latest Mistral Magistral Small Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/magistral-small-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `prompt_mode` (enum, Reasoning) — Enables Mistral's reasoning system prompt; leave unset to disable the default reasoning behavior. [values: "reasoning"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-14b-2512 Mistral Ministral 14B 2512 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-14b-2512.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-14b-latest Mistral Ministral 14B Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-14b-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-3b-2512 Mistral Ministral 3B 2512 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-3b-2512.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-3b-latest Mistral Ministral 3B Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-3b-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-8b-2512 Mistral Ministral 8B 2512 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-8b-2512.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/ministral-8b-latest Mistral Ministral 8B Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/ministral-8b-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-large-2512 Mistral Large 2512 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-large-2512.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-large-latest Mistral Large Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-large-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-medium-3 Mistral Medium 3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-medium-3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-medium-3.5 Mistral Medium 3.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-medium-3.5.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-medium-latest Mistral Medium Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-medium-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-small-2603 Mistral Small 2603 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-small-2603.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/mistral-small-latest Mistral Small Latest on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/mistral-small-latest.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ### mistral/open-mistral-nemo Mistral NeMo on Chat Completions · JSON: https://modelparams.dev/api/v1/models/mistral/open-mistral-nemo.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `stop` (string, Length) — Stops generation when this string is detected. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.5, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0.01, max 1, step 0.01] - `random_seed` (integer, Sampling) — Seed used for deterministic sampling when reproducible outputs are desired. [range: min 0] - `presence_penalty` (number, Sampling) — Penalizes repeated words or phrases to encourage a wider variety of generated content. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes words based on how often they already appear in the generated text. [default: 0] [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON mode output. [default: "text"] [values: "text", "json_object", "json_schema"] - `safe_prompt` (boolean, Metadata) — Controls whether Mistral injects its safety prompt before the conversation. [default: false] ## Fireworks ### fireworks/deepseek-v4-flash-0731 Fireworks DeepSeek v4 Flash 0731 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/deepseek-v4-flash-0731.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/deepseek-v4-flash-vision-exp Fireworks DeepSeek v4 Flash Vision Exp on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/deepseek-v4-flash-vision-exp.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/deepseek-v4-pro Fireworks DeepSeek v4 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/deepseek-v4-pro.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/deepseek-v4-pro-0813 Fireworks DeepSeek v4 Pro 0813 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/deepseek-v4-pro-0813.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/deepseek-v4p1-flash Fireworks DeepSeek V4p1 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/deepseek-v4p1-flash.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/glm-5p2 Fireworks GLM-5p2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/glm-5p2.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/glm-5p3 Fireworks GLM-5p3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/glm-5p3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/glm-5p3-flash Fireworks GLM-5p3 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/glm-5p3-flash.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/gpt-oss-120b Fireworks GPT-OSS 120B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/gpt-oss-120b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/gpt-oss-20b Fireworks GPT-OSS 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/gpt-oss-20b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/inkling Fireworks Inkling on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/inkling.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/kimi-k2p6 Fireworks Kimi K2p6 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/kimi-k2p6.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/kimi-k2p7-code Fireworks Kimi K2p7 Code on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/kimi-k2p7-code.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/kimi-k3 Fireworks Kimi K3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/kimi-k3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/minimax-m2p7 Fireworks MiniMax M2p7 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/minimax-m2p7.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/minimax-m3 Fireworks MiniMax M3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/minimax-m3.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/muse-glimmer-30b Fireworks Muse Glimmer 30B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/muse-glimmer-30b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/qwen3p7-plus Fireworks Qwen3p7 Plus on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/qwen3p7-plus.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/qwen3p8-2p4t-a95b Fireworks Qwen3p8 2p4t A95B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/qwen3p8-2p4t-a95b.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### fireworks/qwen3p8-max Fireworks Qwen3p8 Max on Chat Completions · JSON: https://modelparams.dev/api/v1/models/fireworks/qwen3p8-max.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1, max 100] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ## MiniMax ### minimax/minimax-m2 MiniMax M2 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/MiniMax-M2-subscription MiniMax M2 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/minimax-m2.1 MiniMax M2.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.1.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/minimax-m2.1-highspeed MiniMax M2.1 Highspeed on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.1-highspeed.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/MiniMax-M2.1-highspeed-subscription MiniMax M2.1 Highspeed (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.1-highspeed-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/MiniMax-M2.1-subscription MiniMax M2.1 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.1-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/minimax-m2.5 MiniMax M2.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.5.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/minimax-m2.5-highspeed MiniMax M2.5 Highspeed on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.5-highspeed.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/MiniMax-M2.5-highspeed-subscription MiniMax M2.5 Highspeed (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.5-highspeed-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/MiniMax-M2.5-subscription MiniMax M2.5 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.5-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/minimax-m2.7 MiniMax M2.7 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.7.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/minimax-m2.7-highspeed MiniMax M2.7 Highspeed on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m2.7-highspeed.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/MiniMax-M2.7-highspeed-subscription MiniMax M2.7 Highspeed (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.7-highspeed-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/MiniMax-M2.7-subscription MiniMax M2.7 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M2.7-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ### minimax/minimax-m3 MiniMax M3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/minimax-m3.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] - `reasoning_split` (boolean, Reasoning) — Returns the model's reasoning in a separate reasoning_details field instead of inline with the response. [default: false] ### minimax/MiniMax-M3-subscription MiniMax M3 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/minimax/MiniMax-M3-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Values must be greater than 0 and at most 1. [default: 1] [range: min 0.01, max 1, step 0.01] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0.01, max 1, step 0.01] ## xAI ### xai/grok-4.20-0309-non-reasoning xAI Grok 4.20 0309 Non Reasoning on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.20-0309-non-reasoning.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. xAI accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.20-0309-non-reasoning-subscription xAI Grok 4.20 0309 Non Reasoning (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.20-0309-non-reasoning-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. xAI accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.20-0309-reasoning xAI Grok 4.20 0309 Reasoning on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.20-0309-reasoning.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.20-0309-reasoning-subscription xAI Grok 4.20 0309 Reasoning (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.20-0309-reasoning-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.20-multi-agent-0309 xAI Grok 4.20 Multi Agent 0309 on Responses API · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.20-multi-agent-0309.json - `max_output_tokens` (integer, Length) — Upper bound for output tokens generated in the Responses API response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.7] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `reasoning.effort` (enum, Reasoning) — Controls whether the Responses API request uses the 4-agent or 16-agent multi-agent setup. [values: "low", "medium", "high", "xhigh"] - `text.format.type` (enum, Output) — Controls whether the Responses API returns free-form text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.3 xAI Grok 4.3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.3.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Set to none for non-reasoning requests. [default: "low"] [values: "none", "low", "medium", "high"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.3-subscription xAI Grok 4.3 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.3-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Set to none for non-reasoning requests. [default: "low"] [values: "none", "low", "medium", "high"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.5 xAI Grok 4.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.5.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Defaults to high. Reasoning cannot be disabled. [default: "high"] [values: "low", "medium", "high"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.5-subscription xAI Grok 4.5 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.5-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Defaults to high. Reasoning cannot be disabled. [default: "high"] [values: "low", "medium", "high"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.6 xAI Grok 4.6 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.6.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. xAI accepts up to four stop sequences. grok-4.6 rejects stop because reasoning cannot be disabled. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Defaults to high. Reasoning cannot be disabled. xhigh is grok-4.6 and later. [default: "high"] [values: "low", "medium", "high", "xhigh"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-4.6-subscription xAI Grok 4.6 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-4.6-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Grok performs before responding. Defaults to high. Reasoning cannot be disabled. xhigh is grok-4.6 and later. [default: "high"] [values: "low", "medium", "high", "xhigh"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-build-0.1 xAI Grok Build 0.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-build-0.1.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ### xai/grok-build-0.1-subscription xAI Grok Build 0.1 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xai/grok-build-0.1-subscription.json - `max_completion_tokens` (integer, Length) — Upper bound for visible output tokens generated in the chat completion. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON mode output, or structured JSON schema output. [default: "text"] [values: "text", "json_object", "json_schema"] ## Moonshot AI ### moonshot/kimi-k2.5 Moonshot AI Kimi K2.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.5.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `thinking.type` (enum, Reasoning) — Controls whether Kimi reasons step by step before answering. Thinking is enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide and none blocks tool calls. Moonshot rejects required while thinking is enabled, which is the default on this model. [default: "auto"] [values: "auto", "none"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/kimi-k2.6 Moonshot AI Kimi K2.6 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.6.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `thinking.type` (enum, Reasoning) — Controls whether Kimi reasons step by step before answering. Thinking is enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `thinking.keep` (enum, Reasoning) — Set to all to carry reasoning content from earlier assistant turns into the request; leave null to send only the final answers. [default: null] [values: "all", null] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide and none blocks tool calls. Moonshot rejects required while thinking is enabled, which is the default on this model. [default: "auto"] [values: "auto", "none"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/kimi-k2.6-subscription Moonshot AI Kimi K2.6 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.6-subscription.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `thinking.type` (enum, Reasoning) — Controls whether Kimi reasons step by step before answering. Thinking is enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `thinking.keep` (enum, Reasoning) — Set to all to carry reasoning content from earlier assistant turns into the request; leave null to send only the final answers. [default: null] [values: "all", null] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### moonshot/kimi-k2.7-code Moonshot AI Kimi K2.7 Code on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.7-code.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion, including reasoning tokens. [default: 32768] [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `thinking.type` (enum, Reasoning) — Thinking is always on for this model. Enabled is the only accepted value; disabled returns an error. [default: "enabled"] [values: "enabled"] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide and none blocks tool calls. Moonshot rejects required while thinking is enabled, which is the default on this model. [default: "auto"] [values: "auto", "none"] - `response_format.type` (enum, Output) — Forces the response into plain text, a JSON object, or JSON matching a provided schema. [default: "text"] [values: "text", "json_object", "json_schema"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/kimi-k2.7-code-highspeed Moonshot AI Kimi K2.7 Code Highspeed on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.7-code-highspeed.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion, including reasoning tokens. [default: 32768] [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `thinking.type` (enum, Reasoning) — Thinking is always on for this model. Enabled is the only accepted value; disabled returns an error. [default: "enabled"] [values: "enabled"] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide and none blocks tool calls. Moonshot rejects required while thinking is enabled, which is the default on this model. [default: "auto"] [values: "auto", "none"] - `response_format.type` (enum, Output) — Forces the response into plain text, a JSON object, or JSON matching a provided schema. [default: "text"] [values: "text", "json_object", "json_schema"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/kimi-k2.7-code-highspeed-subscription Moonshot AI Kimi K2.7 Code Highspeed (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.7-code-highspeed-subscription.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion, covering both thinking and the final answer. [range: min 1] - `thinking.type` (enum, Reasoning) — Thinking is always on for this model. Enabled is the only accepted value; disabled returns an error. [default: "enabled"] [values: "enabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### moonshot/kimi-k2.7-code-subscription Moonshot AI Kimi K2.7 Code (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k2.7-code-subscription.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion, covering both thinking and the final answer. [range: min 1] - `thinking.type` (enum, Reasoning) — Thinking is always on for this model. Enabled is the only accepted value; disabled returns an error. [default: "enabled"] [values: "enabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### moonshot/kimi-k3 Moonshot AI Kimi K3 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/kimi-k3.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `thinking.type` (enum, Reasoning) — Controls whether Kimi reasons step by step before answering, or responds directly when set to disabled. [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning Kimi performs before answering. Thinking is always on for this model, so it cannot be turned off. [default: "max"] [values: "low", "high", "max"] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide, none blocks tool calls, and required forces one. [default: "auto"] [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Forces the response into plain text, a JSON object, or JSON matching a provided schema. [default: "text"] [values: "text", "json_object", "json_schema"] ### moonshot/moonshot-v1-128k Moonshot AI Moonshot v1 128K on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/moonshot-v1-128k.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `n` (integer, Length) — How many chat completion choices to generate for the request. [default: 1] [range: min 1, max 5] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, max 1, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to talk about new topics. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide, none blocks tool calls, and required forces one. [default: "auto"] [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/moonshot-v1-32k Moonshot AI Moonshot v1 32K on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/moonshot-v1-32k.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `n` (integer, Length) — How many chat completion choices to generate for the request. [default: 1] [range: min 1, max 5] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, max 1, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to talk about new topics. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide, none blocks tool calls, and required forces one. [default: "auto"] [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ### moonshot/moonshot-v1-8k Moonshot AI Moonshot v1 8K on Chat Completions · JSON: https://modelparams.dev/api/v1/models/moonshot/moonshot-v1-8k.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate in the chat completion. [range: min 1] - `stop` (string, Length) — Stops generation when this sequence is produced. Moonshot accepts up to 5 sequences of at most 32 bytes each. - `n` (integer, Length) — How many chat completion choices to generate for the request. [default: 1] [range: min 1, max 5] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, max 1, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to talk about new topics. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `tool_choice` (enum, Tools) — Controls tool calling: auto lets the model decide, none blocks tool calls, and required forces one. [default: "auto"] [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `top_logprobs` (integer, Observability) — Number of most likely tokens to return a log probability for at each position. [range: min 0, max 20] [only when logprobs = true] ## Vertex ### vertex/gemini-2.5-flash Vertex Gemini 2.5 Flash on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-2.5-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-2.5-flash-lite Vertex Gemini 2.5 Flash Lite on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-2.5-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-2.5-pro Vertex Gemini 2.5 Pro on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-2.5-pro.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3-flash-preview Vertex Gemini 3 Flash Preview on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3-flash-preview.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.1-flash-lite Vertex Gemini 3.1 Flash Lite on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.1-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.1-pro-preview Vertex Gemini 3.1 Pro Preview on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.1-pro-preview.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.1-pro-preview-customtools Vertex Gemini 3.1 Pro Preview Customtools on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.1-pro-preview-customtools.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.5-flash Vertex Gemini 3.5 Flash on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.5-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.presencePenalty` (number, Sampling) — Penalises tokens that already appeared, pushing the model toward new topics. [range: min -2, max 1] - `generationConfig.frequencyPenalty` (number, Sampling) — Penalises tokens in proportion to how often they have already appeared. [range: min -2, max 1] - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.5 Flash reasoning effort. [default: "medium"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.5-flash-lite Vertex Gemini 3.5 Flash Lite on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.5-flash-lite.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.5 Flash reasoning effort. [default: "medium"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.6-flash Vertex Gemini 3.6 Flash on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.6-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ### vertex/gemini-3.7-flash Vertex Gemini 3.7 Flash on Vertex generate content · JSON: https://modelparams.dev/api/v1/models/vertex/gemini-3.7-flash.json - `generationConfig.maxOutputTokens` (integer, Length) — Maximum number of tokens to include in a response candidate. [range: min 1, max 65536] - `generationConfig.candidateCount` (integer, Length) — How many independent completions to generate for one request. [range: min 1, max 8] - `generationConfig.stopSequences` (string, Length) — A list of strings where the model stops generating further tokens. - `generationConfig.temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `generationConfig.topP` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.95] [range: min 0, max 1, step 0.01] - `generationConfig.topK` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [default: 64] [range: min 1] - `generationConfig.seed` (integer, Sampling) — Optional seed used for decoding when reproducible sampling is desired. - `generationConfig.thinkingConfig.thinkingBudget` (integer, Reasoning) — Token budget the model may spend on internal reasoning before answering. 0 disables thinking. Cannot be combined with thinkingLevel. [range: min 0] [except when generationConfig.thinkingConfig.thinkingLevel ≠ null] - `generationConfig.thinkingConfig.includeThoughts` (boolean, Reasoning) — Controls whether Gemini returns available thought summaries in the response parts. [default: false] - `generationConfig.thinkingConfig.thinkingLevel` (enum, Reasoning) — Controls Gemini 3.1 Flash-Lite reasoning effort. [default: "minimal"] [values: "minimal", "low", "medium", "high"] [except when generationConfig.thinkingConfig.thinkingBudget ≠ null] - `generationConfig.responseMimeType` (enum, Output) — MIME type for generated text candidates. [default: "text/plain"] [values: "text/plain", "application/json"] ## Cohere ### cohere/command-a-03-2025 Cohere Command A 03 2025 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-a-03-2025.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ### cohere/command-a-plus-05-2026 Cohere Command A Plus 05 2026 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-a-plus-05-2026.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ### cohere/command-a-reasoning-08-2025 Cohere Command A Reasoning 08 2025 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-a-reasoning-08-2025.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `thinking.type` (enum, Reasoning) — Controls whether the model reasons step by step before producing its final answer. [default: "disabled"] [values: "enabled", "disabled"] - `thinking.token_budget` (integer, Reasoning) — Maximum number of tokens the model may spend on reasoning before answering. [range: min 1] [only when thinking.type = "enabled"] - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ### cohere/command-a-translate-08-2025 Cohere Command A Translate 08 2025 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-a-translate-08-2025.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ### cohere/command-a-vision-07-2025 Cohere Command A Vision 07 2025 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-a-vision-07-2025.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ### cohere/command-r-08-2024 Cohere Command R 08 2024 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-r-08-2024.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT", "OFF"] ### cohere/command-r-plus-08-2024 Cohere Command R Plus 08 2024 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-r-plus-08-2024.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT", "OFF"] ### cohere/command-r7b-12-2024 Cohere Command R7B 12 2024 on Cohere Chat · JSON: https://modelparams.dev/api/v1/models/cohere/command-r7b-12-2024.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop_sequences` (string, Length) — Stops generation when one of these sequences is detected; up to five are allowed. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 0.3] [range: min 0, step 0.1] - `p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 0.75] [range: min 0.01, max 0.99, step 0.01] - `k` (integer, Sampling) — Limits sampling to the K most likely tokens; 0 disables top-k sampling. [default: 0] [range: min 0, max 500] - `frequency_penalty` (number, Sampling) — Penalizes tokens proportional to how often they have already appeared to reduce repetition. [default: 0] [range: min 0, max 1, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [default: 0] [range: min 0, max 1, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `tool_choice` (enum, Tools) — Forces the model to either call a tool or skip tool calls for this request. [values: "REQUIRED", "NONE"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or JSON object output. [default: "text"] [values: "text", "json_object"] - `logprobs` (boolean, Observability) — Controls whether the response includes log probabilities for the generated tokens. [default: false] - `safety_mode` (enum, Metadata) — Controls Cohere's built-in safety instructions applied to the generation. [default: "CONTEXTUAL"] [values: "CONTEXTUAL", "STRICT"] ## Opencode Go ### opencode-go/deepseek-v4-flash-vision-exp-subscription Opencode Go DeepSeek v4 Flash Vision Exp (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/deepseek-v4-flash-vision-exp-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0.01, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/glm-5.3-flash-subscription Opencode Go GLM-5.3 Flash (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/glm-5.3-flash-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/glm-5.3-subscription Opencode Go GLM-5.3 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/glm-5.3-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/grok-4.6-subscription Opencode Go Grok 4.6 (Subscription) on Responses API · JSON: https://modelparams.dev/api/v1/models/opencode-go/grok-4.6-subscription.json - `max_output_tokens` (integer, Length) — Maximum number of tokens to generate in the response, including any hidden reasoning tokens. [range: min 16] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0.01, max 1, step 0.01] ### opencode-go/hy4-preview-subscription Opencode Go Hy4 Preview (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/hy4-preview-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/longcat-2.0-subscription Opencode Go Longcat 2.0 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/longcat-2.0-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0.01, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/qwen3.8-flash-subscription Opencode Go Qwen3.8 Flash (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/qwen3.8-flash-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ### opencode-go/qwen3.8-max-subscription Opencode Go Qwen3.8 Max (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/opencode-go/qwen3.8-max-subscription.json - `max_tokens` (integer, Length) — Maximum number of tokens to generate in the response. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 1.9, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `top_k` (integer, Sampling) — Limits token sampling to the top K most likely next tokens. [range: min 1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared to encourage a wider variety of content. [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have already appeared. [range: min -2, max 2, step 0.1] - `response_format.type` (enum, Output) — Choose plain text or force valid JSON output. [values: "text", "json_object"] ## Groq ### groq/gpt-oss-120b Groq GPT-OSS 120B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/gpt-oss-120b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### groq/gpt-oss-20b Groq GPT-OSS 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/gpt-oss-20b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### groq/gpt-oss-safeguard-20b Groq GPT-OSS Safeguard 20B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/gpt-oss-safeguard-20b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### groq/qwen3-32b Groq Qwen3 32B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/qwen3-32b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### groq/qwen3.6-27b Groq Qwen3.6 27B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/qwen3.6-27b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### groq/qwen3.8-27b Groq Qwen3.8 27B on Chat Completions · JSON: https://modelparams.dev/api/v1/models/groq/qwen3.8-27b.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Groq accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls whether the model reasons before answering. 'none' disables reasoning; 'default' lets the model reason. [default: "default"] [values: "none", "default"] - `reasoning_format` (enum, Reasoning) — Controls how reasoning tokens are returned — hidden from the response, raw within the content, or parsed into a separate field. [values: "hidden", "raw", "parsed"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ## Meta ### meta/Llama-3.3-70B-Instruct Meta Llama 3.3 70B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/Llama-3.3-70B-Instruct.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. - `repetition_penalty` (number, Sampling) — Penalizes tokens that have already appeared to reduce repetition in the output. - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### meta/Llama-3.3-8B-Instruct Meta Llama 3.3 8B Instruct on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/Llama-3.3-8B-Instruct.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. - `repetition_penalty` (number, Sampling) — Penalizes tokens that have already appeared to reduce repetition in the output. - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### meta/Llama-4-Maverick-17B-128E-Instruct-FP8 Meta Llama 4 Maverick 17B 128E Instruct FP8 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/Llama-4-Maverick-17B-128E-Instruct-FP8.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. - `repetition_penalty` (number, Sampling) — Penalizes tokens that have already appeared to reduce repetition in the output. - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### meta/Llama-4-Scout-17B-16E-Instruct-FP8 Meta Llama 4 Scout 17B 16E Instruct FP8 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/Llama-4-Scout-17B-16E-Instruct-FP8.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. - `top_k` (integer, Sampling) — Limits generation to the selected number of highest-probability tokens. - `repetition_penalty` (number, Sampling) — Penalizes tokens that have already appeared to reduce repetition in the output. - `tool_choice` (enum, Tools) — Controls whether the model may call tools, must call one, or skips tool calls. [values: "auto", "none", "required"] - `response_format.type` (enum, Output) — Controls whether the model returns normal text or a schema-constrained JSON object. [default: "text"] [values: "text", "json_schema"] ### meta/muse-spark-1.1 Meta Muse Spark 1.1 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/muse-spark-1.1.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness; use this or Top P, but not both. [default: 1] [range: min 0, max 2, step 0.1] [except when top_p ≠ 1] - `top_p` (number, Sampling) — Controls nucleus sampling; use this or Temperature, but not both. [default: 1] [range: min 0.01, max 1, step 0.01] [except when temperature ≠ 1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have appeared, reducing repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Requests best-effort deterministic sampling for repeated requests. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [values: "minimal", "low", "medium", "high", "xhigh"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON, or schema-constrained JSON. [default: "text"] [values: "text", "json_object", "json_schema"] - `prompt_cache_retention` (enum, Metadata) — Controls whether the prompt cache stays in memory or persists for up to 24 hours. [values: "in_memory", "24h"] ### meta/muse-spark-1.2-contributor Meta Muse Spark 1.2 Contributor on Chat Completions · JSON: https://modelparams.dev/api/v1/models/meta/muse-spark-1.2-contributor.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness; use this or Top P, but not both. [default: 1] [range: min 0, max 2, step 0.1] [except when top_p ≠ 1] - `top_p` (number, Sampling) — Controls nucleus sampling; use this or Temperature, but not both. [default: 1] [range: min 0.01, max 1, step 0.01] [except when temperature ≠ 1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have appeared, reducing repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Requests best-effort deterministic sampling for repeated requests. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model should perform before producing an answer. [values: "minimal", "low", "medium", "high", "xhigh"] - `response_format.type` (enum, Output) — Controls whether the model returns text, JSON, or schema-constrained JSON. [default: "text"] [values: "text", "json_object", "json_schema"] - `prompt_cache_retention` (enum, Metadata) — Controls whether the prompt cache stays in memory or persists for up to 24 hours. [values: "in_memory", "24h"] ## DeepSeek ### deepseek/deepseek-chat DeepSeek Chat on Chat Completions · JSON: https://modelparams.dev/api/v1/models/deepseek/deepseek-chat.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether DeepSeek uses thinking mode before producing the final answer. [default: "disabled"] [values: "disabled", "enabled"] ### deepseek/deepseek-reasoner DeepSeek Reasoner on Chat Completions · JSON: https://modelparams.dev/api/v1/models/deepseek/deepseek-reasoner.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether DeepSeek uses thinking mode before producing the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls DeepSeek thinking effort when thinking mode is enabled. [default: "high"] [values: "high", "max"] [only when thinking.type = "enabled"] ### deepseek/deepseek-v4-flash DeepSeek-V4 Flash on Chat Completions · JSON: https://modelparams.dev/api/v1/models/deepseek/deepseek-v4-flash.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether DeepSeek uses thinking mode before producing the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls DeepSeek thinking effort when thinking mode is enabled. For compatibility, low, medium, and xhigh are also accepted, but low and medium are mapped to high, and xhigh is mapped to max. [default: "high"] [values: "high", "max"] [only when thinking.type = "enabled"] ### deepseek/deepseek-v4-pro DeepSeek-V4 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/deepseek/deepseek-v4-pro.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [default: 4096] [range: min 1] - `temperature` (number, Sampling) — Controls randomness. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Controls nucleus sampling. In DeepSeek thinking mode this parameter is accepted for compatibility but has no effect. [default: 1] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether DeepSeek uses thinking mode before producing the final answer. [default: "enabled"] [values: "enabled", "disabled"] - `reasoning_effort` (enum, Reasoning) — Controls DeepSeek thinking effort when thinking mode is enabled. [default: "high"] [values: "high", "max"] [only when thinking.type = "enabled"] ## Perplexity ### perplexity/sonar Perplexity Sonar on Chat Completions · JSON: https://modelparams.dev/api/v1/models/perplexity/sonar.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 128000] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `search_mode` (enum, Metadata) — Selects the corpus the model searches when grounding its answer. [values: "web", "academic", "sec"] - `search_recency_filter` (enum, Metadata) — Restricts web search results to a recent time window. [values: "hour", "day", "week", "month", "year"] - `search_domain_filter` (string, Metadata) — Limits search to, or excludes, specific domains. - `search_after_date_filter` (string, Metadata) — Restricts search results to content published after this date (MM/DD/YYYY). - `search_before_date_filter` (string, Metadata) — Restricts search results to content published before this date (MM/DD/YYYY). - `web_search_options.search_context_size` (enum, Metadata) — Controls how much web search context is retrieved before generating the answer. [default: "low"] [values: "low", "medium", "high"] - `return_images` (boolean, Metadata) — Controls whether the response may include related images from the search. [default: false] - `return_related_questions` (boolean, Metadata) — Controls whether the response includes suggested follow-up questions. [default: false] - `disable_search` (boolean, Metadata) — Turns off web search so the model answers from its own knowledge only. [default: false] ### perplexity/sonar-deep-research Perplexity Sonar Deep Research on Chat Completions · JSON: https://modelparams.dev/api/v1/models/perplexity/sonar-deep-research.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 128000] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning and searching the model performs before producing the report. [values: "minimal", "low", "medium", "high"] - `search_mode` (enum, Metadata) — Selects the corpus the model searches when grounding its answer. [values: "web", "academic", "sec"] - `search_recency_filter` (enum, Metadata) — Restricts web search results to a recent time window. [values: "hour", "day", "week", "month", "year"] - `search_domain_filter` (string, Metadata) — Limits search to, or excludes, specific domains. - `search_after_date_filter` (string, Metadata) — Restricts search results to content published after this date (MM/DD/YYYY). - `search_before_date_filter` (string, Metadata) — Restricts search results to content published before this date (MM/DD/YYYY). - `web_search_options.search_context_size` (enum, Metadata) — Controls how much web search context is retrieved before generating the answer. [default: "low"] [values: "low", "medium", "high"] - `return_images` (boolean, Metadata) — Controls whether the response may include related images from the search. [default: false] - `return_related_questions` (boolean, Metadata) — Controls whether the response includes suggested follow-up questions. [default: false] ### perplexity/sonar-pro Perplexity Sonar Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/perplexity/sonar-pro.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 128000] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `search_mode` (enum, Metadata) — Selects the corpus the model searches when grounding its answer. [values: "web", "academic", "sec"] - `search_recency_filter` (enum, Metadata) — Restricts web search results to a recent time window. [values: "hour", "day", "week", "month", "year"] - `search_domain_filter` (string, Metadata) — Limits search to, or excludes, specific domains. - `search_after_date_filter` (string, Metadata) — Restricts search results to content published after this date (MM/DD/YYYY). - `search_before_date_filter` (string, Metadata) — Restricts search results to content published before this date (MM/DD/YYYY). - `web_search_options.search_context_size` (enum, Metadata) — Controls how much web search context is retrieved before generating the answer. [default: "low"] [values: "low", "medium", "high"] - `return_images` (boolean, Metadata) — Controls whether the response may include related images from the search. [default: false] - `return_related_questions` (boolean, Metadata) — Controls whether the response includes suggested follow-up questions. [default: false] - `disable_search` (boolean, Metadata) — Turns off web search so the model answers from its own knowledge only. [default: false] ### perplexity/sonar-reasoning-pro Perplexity Sonar Reasoning Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/perplexity/sonar-reasoning-pro.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1, max 128000] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `search_mode` (enum, Metadata) — Selects the corpus the model searches when grounding its answer. [values: "web", "academic", "sec"] - `search_recency_filter` (enum, Metadata) — Restricts web search results to a recent time window. [values: "hour", "day", "week", "month", "year"] - `search_domain_filter` (string, Metadata) — Limits search to, or excludes, specific domains. - `search_after_date_filter` (string, Metadata) — Restricts search results to content published after this date (MM/DD/YYYY). - `search_before_date_filter` (string, Metadata) — Restricts search results to content published before this date (MM/DD/YYYY). - `web_search_options.search_context_size` (enum, Metadata) — Controls how much web search context is retrieved before generating the answer. [default: "low"] [values: "low", "medium", "high"] - `return_images` (boolean, Metadata) — Controls whether the response may include related images from the search. [default: false] - `return_related_questions` (boolean, Metadata) — Controls whether the response includes suggested follow-up questions. [default: false] - `disable_search` (boolean, Metadata) — Turns off web search so the model answers from its own knowledge only. [default: false] ## Xiaomi ### xiaomi/mimo-v2.5 Xiaomi MiMo v2.5 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xiaomi/mimo-v2.5.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, covering both the thinking trace and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values are more focused; higher values are more varied. Ignored while thinking is enabled, where it is forced to 1.0. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Nucleus sampling cutoff. Ignored while thinking is enabled, where it is forced to 0.95. [default: 0.95] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether MiMo reasons step by step before answering. Enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### xiaomi/mimo-v2.5-pro Xiaomi MiMo v2.5 Pro on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xiaomi/mimo-v2.5-pro.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, covering both the thinking trace and the final answer. [range: min 1] - `stop` (string, Length) — Up to a few sequences where generation stops; the stop text is not included in the output. - `temperature` (number, Sampling) — Controls randomness. Lower values are more focused; higher values are more varied. Ignored while thinking is enabled, where it is forced to 1.0. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Nucleus sampling cutoff. Ignored while thinking is enabled, where it is forced to 0.95. [default: 0.95] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `frequency_penalty` (number, Sampling) — Penalizes tokens in proportion to how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `thinking.type` (enum, Reasoning) — Controls whether MiMo reasons step by step before answering. Enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ### xiaomi/mimo-v2.5-subscription Xiaomi MiMo v2.5 (Subscription) on Chat Completions · JSON: https://modelparams.dev/api/v1/models/xiaomi/mimo-v2.5-subscription.json - `max_completion_tokens` (integer, Length) — Maximum number of tokens to generate, covering both the thinking trace and the final answer. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values are more focused; higher values are more varied. Ignored while thinking is enabled, where it is forced to 1.0. [default: 1] [range: min 0, max 2, step 0.1] [except when thinking.type = "enabled"] - `top_p` (number, Sampling) — Nucleus sampling cutoff. Ignored while thinking is enabled, where it is forced to 0.95. [default: 0.95] [range: min 0, max 1, step 0.01] [except when thinking.type = "enabled"] - `thinking.type` (enum, Reasoning) — Controls whether MiMo reasons step by step before answering. Enabled by default; set disabled to respond directly. [default: "enabled"] [values: "enabled", "disabled"] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ## Cerebras ### cerebras/zai-glm-4.7 Cerebras Zai GLM-4.7 on Chat Completions · JSON: https://modelparams.dev/api/v1/models/cerebras/zai-glm-4.7.json - `max_completion_tokens` (integer, Length) — Maximum number of output tokens the model may generate, including reasoning tokens. [range: min 1] - `stop` (string, Length) — A string or list of strings where the API will stop generating further tokens. Cerebras accepts up to four stop sequences. - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. Adjust this or top_p, not both. [range: min 0, max 2, step 0.1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [range: min 0, max 1, step 0.01] - `frequency_penalty` (number, Sampling) — Penalizes tokens by how often they have appeared, reducing verbatim repetition. [default: 0] [range: min -2, max 2, step 0.1] - `presence_penalty` (number, Sampling) — Penalizes tokens that have already appeared, encouraging the model to introduce new topics. [default: 0] [range: min -2, max 2, step 0.1] - `seed` (integer, Sampling) — Seed used for best-effort deterministic sampling when reproducible outputs are desired. - `reasoning_effort` (enum, Reasoning) — Controls how much reasoning the model performs before answering. 'none' disables reasoning. [values: "none", "low", "medium", "high"] - `clear_thinking` (boolean, Reasoning) — When true, the model's thinking from previous turns is excluded from the conversation context; when false, it is preserved, which is useful for agentic workflows. [default: true] - `response_format.type` (enum, Output) — Forces the response into plain text or a JSON object. [default: "text"] [values: "text", "json_object"] ## Thinking Machines ### thinking-machines/Inkling Thinking Machines Inkling on Chat Completions · JSON: https://modelparams.dev/api/v1/models/thinking-machines/Inkling.json - `max_tokens` (integer, Length) — Maximum number of output tokens the model may generate. [range: min 1] - `temperature` (number, Sampling) — Controls randomness. Lower values make outputs more focused; higher values make them more varied. [default: 1] - `top_p` (number, Sampling) — Controls nucleus sampling by limiting generation to tokens within the selected cumulative probability. [default: 1] [range: min 0, max 1] - `reasoning_effort` (enum, Reasoning) — Controls how much thinking Inkling performs before answering. Accepts a preset name or a number between 0.0 and 0.99; presets map to numeric effort levels (none=0.0, minimal=0.1, low=0.2, medium=0.7, high=0.9, xhigh=0.99). [default: "high"] [values: "none", "minimal", "low", "medium", "high", "xhigh"] - `separate_reasoning` (boolean, Reasoning) — Returns the model's reasoning in a dedicated reasoning_content field instead of interleaving it with the final message content. [default: true]