Cursor's experience with the Claude 4 sonnet thinking model

A few days ago, when Claude 4 dropped, Cursor hopped on it right away, and after using it for two days, it’s pretty decent!Compared to the claude-3.7-sonnet(thinking) model, here are the immediate pros I’ve noticed (still updating):

1. It gets dev needs better—no more random or repetitive tweaks like 3.7, fewer unnecessary changes, so it feels faster than 3.7;

2. It can think ahead—spot a bug on a page (like missing null checks) and fix similar issues across the board;

3. Less likely to mess up—used it to optimize a performance issue, and the tweaks were spot-on, barely needed any follow-ups;

4. It auto-generates a PERFORMANCE_OPTIMIZATION.md doc (though it misplaced it) detailing the optimization suggestions, process, code changes, and next steps.