Cursor has been bad with this LLM. It shifts to caveman, it’s lazy, its focus is weird, like it’s context isn’t fully loading ,or summarized, so it would miss obvious things that I need to call out and actually think on its behalf? It was not like this previously with the cursor models. I’ve had to go back to chatgpt. I’ve removed my personal coding system prompt thinking thats the problem with no improvement.
Hi @K_B Thank you for the forum post and sorry for the frustration here.
A few things that are probably worth trying:
- Start a fresh chat, particularly if the current one is long.
@-mention the key files or folder so the relevant context is explicit.- Compare the same task using Grok 4.6 and Composer or another model.
If Grok still misses the same details, could you share one concrete example, your Cursor version, and the Request ID from the chat’s ... menu under Copy Request ID? That will give us something specific to investigate.