Description

I've been trying out pydantic AI with an OpenAiResponsesModel gpt-5 agent for a use-case involving function calling and reasoning.

From what I've gathered while poking around the code, it doesn't seem like the
previous_response_id parameter mentioned here in openai's docs is being sent along by pydantic AI?

This page also mentions the parameter, and here there are claims of improved evaluation performance possibly from the use of the parameter (although the text is a little ambiguous on what exactly the improvements measured).

Perhaps agents would benefit from always sending the parameter along when there are previous responses that have been made?

References

No response