Hi everyone, I am a big fan of having alternatives to Sonnet. Sonnet is fantastic but there are many tasks where sonnet is an overkill and I like to use stuff like deepseek for these simpler tasks. Gemini - especiall the 1206 exp and flash thinking - is pretty good at this too and has faster responses usually than deepseek used by cursor, so I would like to use it occasionally but it is so hit and miss that it is a waste of time to use it. Frequently it generates code that cursor straight up can’t apply. I don’t see this issue when using it with aider, where it is able to apply the generated diffs well enough.
Is the cursor team working on better integrating gemini with cursor?
Thanks.
1 Like
+1, usually asking it to apply again works, but it’s annoying and wastes time.
Could you share a screenshot of when Gemini generates code that Cursor can’t apply? This would help us understand what’s happening and improve the integration
I don’t know how I missed seeing this! Here’s a screenshot. This is in python:
Ah, looks like Gemini is not outputting in the correct format that Cursor expects!
This could require some tweaking behind the scenes, so I’ll pass this to the team. For now, I’d recommend just using a different model, like Claude, that works a lot more reliably here.
Yes, I am using different models for now. It has gotten worse in the last two days. Usually once in every few prompts gemini would not output correctly, but in the last two days it hasn’t output a properly formated response once.
