Gpt-5-mini models return empty output_text in the Responses API

API Bugs
1

What’s the issue:

  • gpt-5-mini models default to reasoning mode, resulting in empty output_text

  • Users were experiencing confusion when response.output_text returned empty strings

  • The fix ensures consistent behavior and provides sensible defaults

File being refrenced - src/openai/resources/responses/responses.py

2

nothing is linked.

It seems this could be the pull request you are interested in.

There is no output_text helper content if there is no user-directed AI content produced.

The core issue is setting a max_output_tokens too low for the AI to finish its reasoning, or writing an output parser that expects only one output object in the list, while the first can be a reasoning summary item.

It is also possible to have an AI model that never transitions out of its reasoning container to direct text to the user, but produces a tool call you also are not looking for or which is hallucinated and not passed through the API due to no recipient, or a stop.

1 Like
3

It’s also happening with other models.