Hey, thanks for the report, and for the screenshot with the error and the Request ID, that helped me track it down.
What’s happening: gemini-3.5-flash-lite isn’t one of the officially supported models in Cursor’s model picker. When you add it as a custom model via the OpenAI-compatible routing, the first turn works, but on later turns with tool calls Google returns 400 Function call is missing a thought_signature. The reason is that for flash-lite slugs Cursor currently doesn’t include the required thought signatures when it resends the conversation history that contains tool calls.
This is a known issue, the same thing was reported before for gemini-3.1-flash-lite-preview. We’ve passed it to the team. I can’t share an ETA yet, but I’ll reply here when there’s an update.
For now, the workaround is to use gemini-3.5-flash (supported natively, no custom key needed) or any other non flash-lite Gemini model with your own key. Tool calls work fine there.
