Autogen Support for gpt-5 reasoning

[ravneetgrewal]

Has the OpenAI ext module been fixed for supporting Reasoning configurations and getting reasoning chunks back?

I am able to port the existing code to use GPT-5 Fast (using 'gpt-5-chat-latest' in the api as model-name/id) along with some other modifications like max_completion_token instead of max_tokens.

GPT-5 Fast works alright, and fast, but the thinking models (using 'gpt-5' in the api as model-name/id) is very slow and there are two issues here:

1. I cannot pass the reasoning object so I can set the reasoning to low
2. in the return we will not get the reasoning object that is under Response APIs instead of the Completion APIs....

Anyone with any insight on this for newer thinking models?

Thanks!
R