I often encounter this problem. The reason is that I ask models to help me complete some very complex tasks, such as asking it to help me refactor a 4000-line Javascript code.
It first explained its refactoring plan to me, but before it actually started to change the code, it would keep in the generating state. Until more than ten minutes passed, I heard the prompt sound of the Cursor, and it was still in the generating state.
Even if I create a new chat, change the model, or restart the Cursor, the structure is consistent.
I mainly use claude-4-sonnet (no deep thinking), but even if I change to the thinking model and even turn on Max mode, the situation is the same. I also tried gemini-2.5-pro and o4-mini and o3, all the same.
But I know why the problem happened, it’s because I asked it to do something too complicated. Once I reduced the requirements and asked it to start in steps, it worked properly.