AI Model Parameter Support Checker

Every hosted LLM API accepts a slightly different set of request parameters, and the mismatches are silent until a request 400s. This tool encodes, for each model, whether a given parameter is supported, supported under a different field name, or not supported — plus its valid range, default value, whether it is required, and the recommended alternative when it is rejected. It is a lookup, not a calculation: nothing is sent to any provider, and the same input always returns the same answer.

The data comes straight from the official API references. Three rules do most of the work:

• OpenAI reasoning models (o1, o3, o3-mini, o4-mini, GPT-5 reasoning) reject the sampling parameters entirely — temperature, top_p, presence_penalty, frequency_penalty, logprobs, logit_bias — and require max_completion_tokens in place of max_tokens.
• Anthropic Claude caps temperature at 1.0 (not 2.0), requires max_tokens on every call, renames stop to stop_sequences, makes system a top-level string, and offers no penalty or log-probability fields.
• Google Gemini keeps the same concepts but nests them under generationConfig in camelCase: topP, topK, maxOutputTokens, stopSequences, and responseSchema for structured output.

The porting check compares any two models across all 14 parameters and flags exactly what changes — what breaks, what needs renaming, where a range shrinks, and what becomes required — because that comparison is the real cause of most porting failures. The support matrix renders every one of the 196 model-by-parameter cells at once, so the answer is on-screen before you touch a control. A built-in cross-check confirms all 196 cells resolve against the source table on every build.

• OpenAI — Chat Completions API reference (parameters)
• OpenAI — Reasoning models guide (unsupported params, max_completion_tokens, reasoning_effort)
• Anthropic — Messages API reference (max_tokens, temperature 0–1, top_p/top_k, system, stop_sequences)
• Anthropic — Extended thinking (forces temperature 1, disallows top_p/top_k)
• Google — Gemini generateContent reference (generationConfig fields)

The support table was last cross-checked against these official references on 2026-07-11. It is reviewed whenever a provider ships a new model family or changes a parameter's behaviour. This tool covers capability only — it does not recommend parameter values, decode error codes, or estimate cost; see the related tools below for those.