Same here. I can’t get the model to properly “stop” its current line of thought. Even if I “stop” the model several times, or provide explicit instructions to stop, it wont field any new requests unless it’s finished it’s line of thinking. After it finishes thinking/edits it accepts new requests.
To be clear, the stop button still “stops” the model from literally generating text, but it is acting more like a “pause” button. Started behaving like this today for me.
I have noticed this too, and have flagged the issue to the Cursor team.
If you have any examples, please reply to this chat with:
Request IDs
Your current Cursor version (Cursor → About Cursor → Copy)
What model you are using (this will be in the request ID, but I’d be good to know here so we can observe at a quick glance whether this is specific to Claude)
Personally, I noticed when swapping from Claude → o3 that it followed instructions.
Exactly the same for me. Stop is acting like a pause, as summarised above. I created a separate report before seeing this thread. But problem also exist if you try to override whats going on while Cursor awaits for tool invocation confirmation.
Very painful. have to wait for the whole think-cycle which may include editing, trying to commit etc. And only then Cursor is willing to hand over, and I have to rollback everything
Better for me - I’m able to stop it, provide feedback, and have it respond to that feedback rather than plowing ahead with its last effort before recognizing the input.
Ongoing Service Degradation