Just trying this model for the first time today. It seems smart, but gets stuck in internal debates about what tools it’s going to use.
Can this be solved with simple prompting/skills to make it more Cursor friendly? I gave up and went back to Opus and it was chewing through tokens and seemed to get stuck in a loop unable to use the most basic tools….
If anyone has any tips for giving Gemini 3.1 better tool use skills in Cursor, it would be hugely appreciated!
Hey, this is a known issue with Gemini 3.1. The model often gets stuck “thinking” about which tools to use instead of actually calling them. Another user ran into something similar: an infinite loop after finishing a task.
Right now, Gemini models are worse at tool calling in agent mode than Claude. If you need stable tool use (file edits, terminal, browser), it’s better to switch to Claude Sonnet or Opus.
If you want to keep experimenting with Gemini, you can try:
The team is aware of the loop detection issue for Gemini. Let me know if you need anything else.