Cursor AI seems to be performing a little worse with the introduction of the new Grok LLM model

Discussions
,
1

It’s just a hunch right now, but ever since Grok became the de facto standard, the Rust implementation seems to be producing more (semantic) errors.
I use Rust and Next.js (TypeScript).
Right now, it’s just a hunch that the old Composer LLM model seemed to run better. Has anyone else noticed this?

6

I came here for the same reason. In the past two days, Cursor with AUTO mode have produced a few cases of blatant syntax errors in Python code, something I haven’t really seen Cursor handing off before. Left me a bit puzzled, but too early to tell if it was just flukes, I guess

7

Hey all!

If you have Request IDs (specifically with Share Data enabled), they really help the team look at model behavior and improve it.

You can also always manually select Composer 2.5 if you’re worried about the model behind Auto and want to be sure (or test out different models for different behavior)

8

Unfortunately I don’t, but thanks for the offer to look into it – it helps reinstantiating confidence into the product, and I’ll let you know if I run into a particularly bad session again!