OpenAI 400: "Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead." (also 'temperature' unsupported) with o-series or GPT-5 models

Cause: OpenAI's reasoning models (o1, o3, o4-mini, the GPT-5 family and later) spend part of their output budget on hidden reasoning before writing the visible answer. For them the Chat Completions API accepts max_completion_tokens (which caps reasoning plus visible output) and rejects the older max_tokens. Most reasoning models also reject or ignore sampling settings such as temperature, top_p, presence/frequency penalties and logprobs, with a similar 'Unsupported parameter' or 'Unsupported value' error. Code and frameworks written for gpt-4o-style models hit this when you switch the model name. How to check: the error names the parameter. Search your code and framework config for max_tokens, temperature and top_p being set for that model. Fix: 1. Replace max_tokens with max_completion_tokens in Chat Completions requests to reasoning models, and give it enough room: if the budget is too small, reasoning can use it all and the visible answer comes back empty with finish_reason 'length'. 2. Remove temperature, top_p and similar parameters for reasoning models (or set only the values the model accepts); control effort with reasoning_effort instead. 3. With the Responses API, the equivalent cap is max_output_tokens, and reasoning settings go in reasoning: {effort: ...}. 4. In frameworks (LangChain, LlamaIndex, LiteLLM, Azure OpenAI wrappers), upgrade to a release that maps max_tokens automatically for reasoning models, or set the framework's max_completion_tokens option; on Azure OpenAI use an API version that supports the parameter. After: test with a small prompt and check finish_reason and usage (reasoning tokens) to tune the budget. Sources: https://platform.openai.com/docs/guides/reasoning https://community.openai.com/t/why-was-max-tokens-changed-to-max-completion-tokens/938077 https://github.com/simonw/llm/issues/724