Performance of Gemini 2.5 Pro Models in Cursor

The new Gemini 2.5 pro model has built-in Cursor without going through the API, but when actually using it in agent mode, I feel its performance is not optimal but when I use the API directly on Cline 3.8

Is integration with agent mode in the works for this model?

no, agent only available on o3 and claude sonnet

Not true. It is supported now.

i dont think that you can just trust some tweets.

I tried it. it cant do much, its fast but doesnt use most tools just outputs commands to use but doesnt use them.

Answers are extremely short and inacurate

Any plan to integrate Gemini into the agent functionality? The promise of a 1 MM token context window on such a powerful model in Agent mode is exciting. I hate to see the opportunity lost. Is there a codebase to fork and built out agentic functionality by my team?

“some tweets”
This is the Head of Design of Cursor. Not a random user.
Try to update. For me its working perfectly fine. Close as Agent of Sonnet.
Creates, read files and can think about codebase.
Only image support is lacking for whatever reason cause Gemini support images

Ok interesting. I had just tested fresh.

Did you use detailed prompts? or simple task assignment?

Some simple some detailed.
the only issue for me are the rate limits and the lack of image support.
only this 2 are my difference between Sonnet and Gemini in agent mode

1 Like