I have exactly same problem. After 3 I spent all limits. (after upgrade to 1.2.1).
It just eat to much token in each response. o3 parses 3 small files (300-400 lines each) and took 800k tokens. This is not normal guys.
Old version:
Just checked 1.2.0 version. Same problem. I suppose something changed in Cursor backend. I think it’s just unusable right now
