New SDK's in training sets

Codex
1

Can you please incorporate the new api sdk’s in the models? I am tired of chat completions and temperature being added incorrectly on easy scripts and I have to manually fix the simplest script of having gpt write code to talk to itself incorrectly.

If I ask gpt to write an api, it should know how to write an api to openai, this is basic stuff.

I have to keep uploading the documentation as a google drive file and try 5 different times or do it myself for simple scripts.

Thought for 59s

You passed a Chat Completions–only parameter to the Responses API.

  • The run errors show: TypeError: Responses.create() got an unexpected keyword argument 'response_format'. This is why most files failed.

  • In the SDK doc you attached, Chat Completions is documented and requires messages with model, not input, and is where response_format belongs.

  • You used the Responses endpoint with input and also sent response_format. That mix is invalid for your installed SDK. Use either:

    • Responses API: client.responses.create(model=..., input=[...]) and drop response_format. Read text via resp.output_text or the output array.

    • Chat Completions: client.chat.completions.create(model=..., messages=[...], response_format=...) and read via resp.choices[0].message.content. The doc you shared shows the Chat Completions contract.

Other mismatches to the SDK contract:

  • If you switch to Chat Completions, use max_tokens, not max_output_tokens (that’s Responses).

  • If you stay on Responses, keep input not messages, and omit response_format. The error proves your SDK build rejects it.

Net: pick one API and match its parameters. Your current mix is the root cause.