O3-mini bad at calling tools in general?

I’ve been experimenting with the Agent mode with different models and Ive noticed that especially the models from OpenAI are really bad at using tools.

Sometimes it announces to make edits and nothing happens. Even searching the codebase goes a lot smoother with Anthropic models.

Is this an implementation/prompt thing from Cursor or are OpenAI models generally not optimized for tool use?

1 Like