The coding capability of Gemini 2.5 pro 0325 is amazing. IMO way better than Claude 3.7 especially in debugging, logical implementation. However, it always fell in tool use like reading file, editing file. The error messages are “Model provided invalid arguments for tool call.”
My temporal solution is to use Gemini for complicate task to generate the code in chat first, and use Claude to apply to tool use. But this is definitely critical to get fix from cursor end.
Try to disable “thinking” and run Gemini
For me it still thinks but kind of has no issues applying tools
Also sometimes restarting cursor completely helps
Sometimes with starting a new chat
But lately I was able to actually use Gemini reliably with tools (big progress as about a few weeks back it wasn’t usable outside of chat, not as an agent)
I cant’t use gemini without thinking enabled anymore. if i disable thinking it will auto-enable the “auto” mode. anyone else have this issue? recently that still worked
Hey, this model has only one mode, thinking, which is why it switches you to auto.