Claude 4 Sonnet is really disappointing in Cursor so far

Discussion
1

Claude 4 Sonnet is really disappointing, the most simple things it fails at, like easy front end fixes.

Also:

  • Overthinks just like claude 3.7 (over)thinking does
  • Runs commands to ‘check the code’??
  • Does not retrieve the right context, thinks it knows the right context way too quickly and then just works with what it has retrieved after a quick search
  • Doesn’t reflect, makes fixes but then also introduces new issues and doesn’t realize because it does not reflect on the fixes it did
2

Hi, Its a new model, that means you have to check how does the model work, what is different in it, how the prompts and approaches have to be adjusted and test a bit.

So many people report that it works really great so its likely with some improvements you can get it to work well too.

Here is the link to Claude 4 Prompt best practices

3

Hey, which version of Cursor are you using?

4

0.50.6

5

the issue im talking about is not just model behavior and fixed by prompting unfortunately