Is it me or Grok is one the fastest and most accurate models

Discussions
1

I am new to Cursor and been traying different models, and Grok is fast and accurate, BUT is way too eager to make changes, so need to be constantly constrained with plan and ask modes. But is excellent in coding, still for linguistic and literary capacity Claude Sonnet continues to be the greatest.

2 Likes
2

Grok 4 often performs very poorly because of Cursor. Grok 4 Fast is very lazy and shallow. Grok Code Fast is very good for simple tasks.

2 Likes
3

I think Grok is best

1 Like
4

Any referrings that Grok models works better “outside” from Cursor IDE?

5

I’m just stating that I haven’t been able to use them stably inside the Cursor IDE.

6

I use Grok-4-fast-reasoning as the default model for my "Ask” mode. It’s surprisingly great and consistent.

Grok 4 is slow af so I use GPT-5 or Sonnet 4.5 over it.

1 Like
7

Grok 4.1 exists but not in Cursor

8

As someone who’s been using LLMs for over three years across pretty much every major model, Grok-4 stands out as one of the least reliable. It regularly produces overconfident answers even when its internal data is clearly inconsistent. It also has a habit of merging unrelated details to maintain a narrative rather than re-checking whether the inputs actually match.

It rarely validates its own assumptions, so once it commits to a line of thought it keeps going, even when new information contradicts it. Structured inputs like logs or tables often get mis-parsed, and it treats guessed connections as facts. The model tends to prioritise conversational flow over accuracy, and it doesn’t challenge the user when something doesn’t add up.

If it drifts into an error state, it doesn’t self-correct — it just reinforces the mistake. All of this creates a confident tone that masks weak reasoning. For anyone relying on consistency or multi-step analysis, these issues show up very quickly.

9

I think it’s just you. I don’t mean any disrespect but my mileage varies greatly. Claude, composer-1, Gemini and GPT are much more stable. Sometimes they are not as fast, but I prefer constituency and accuracy over speed.