Temperature parameter included in GPT-5 prompts from SDK

​

Summary

Issue: When using the Braintrust SDK to fetch a prompt configured for a GPT-5 model, the returned parameters include a temperature value that OpenAI rejects, causing errors when you pass the parameters directly to OpenAI’s API. Cause: Whether a GPT-5 model accepts temperature depends on the reasoning effort. GPT-5.1 and later accept temperature only when reasoning effort is set to none. At any higher reasoning effort, and for older GPT-5 models (gpt-5, gpt-5-mini, gpt-5-nano) and GPT-5 Pro, OpenAI rejects temperature. Braintrust removes temperature from a prompt’s stored parameters when the selected model and reasoning effort don’t support it, but a prompt saved before this behavior can still carry a temperature value that OpenAI rejects. Resolution: Re-save the prompt in the Braintrust prompt editor so its stored parameters are sanitized for the selected model and reasoning effort. If you build requests to OpenAI directly, remove temperature from the fetched parameters when the model doesn’t support it.

​

Resolution steps

​

Option 1: Re-save the prompt (recommended)

​

Step 1: Open and save the prompt in the editor

Open the prompt in the Braintrust prompt editor and save it. Saving sanitizes the stored parameters for the selected model and reasoning effort, removing temperature when the model doesn’t accept it. The SDK then returns parameters that OpenAI accepts.

​

Option 2: Filter the parameter before calling OpenAI

​

Step 1: Remove temperature from the fetched parameters

If you call OpenAI directly with a prompt that still includes an unsupported temperature, remove it from the fetched parameters before making the request. As a rule, keep temperature for GPT-5.1 and later only when reasoning_effort is none, and remove it for older GPT-5 models and GPT-5 Pro. See GPT-5 prompts include unsupported temperature for a TypeScript example.