As I told before you could burn your token in during of the day. And when you working by Auto regime, it could send your request to gemini fast and you see the same loop of thinking which does not stop and going to burn your tokens. Therefore it's better to test some models by self before come to Cursor.
And yes, I understand that Cursor can't to share the same level of tokens as other AI providers. But anyway even 20/60$ subs does not align to my consume and 200$ is expensive for me.