Is grok really as good as they claim ? Have you tried it?
I felt claude was better at code gneration grok was better at optimization but failed to generate useful code.
meh…strange but 3.5 is better then 3.7
Is claude-3.7-sonnet-thinking handled as a regular premium model? I wish the documentation for the available models, their capabilities and pricing (Cursor – Models) was updated as soon as new models are added.
oh def. not
Answering my own question: It seems that claude-3.7-sonnet-thinking just uses 1 premium/fast request, same as (most) other premium models.
If Cursor implements Claude 3.7, it would cost them a lot. Claude claimed that 3.7’s pricing is the same as 3.5, however in 3.7 thinking also consumes the output token, which is $15 per mil. I tried their Claude Coder (with Claude 3.7) and by simply /init a demo react/next.js example project without doing any coding. It already costs $0.20 
1 Like
You can’t change mid conversation
it was released now
It seems that the plan for the future is to allow us to opt for harder thinking mode, at a cost of 3x fast requests.
https://x.com/sualehasif996/status/1894094715479548273
1 Like
im feeling the agi
also claude code is amazing
1 Like
I was just about to suggest this.
Hopefully there’s an something in between 1-3 requests.
AFIAK, the thinking token count is fully variable in the API so ideally we could use a slider that snaps to number of requests in 0.5 increments to find our own sweet spot.
If more thinking one-shots a solution (using 2 requests), that’s better than 4 requests/replies with standard thinking, or one-shot on max thinking (also using 3 requests.) 
1 Like
After testing my own project in Cursor, I found that Cloud 3.7’s thought process is extremely fast and confident.
3.7 feels like an improvement for me (except when it was losing connection earlier today and chewing up 3-7x as many tokens as necessary). I had a nextjs 15.1.7 issue with dynamic routing and NextAuth 5 beta that claude 3.5 choked on and went in circles while wrecking code. 3.7 fixed it quickly with a creative solution that worked perfectly.
3 Likes
Can’t say better, very frustrating
3.7 seems exceptionally fluent for me. No more dumb implementations and understands the code better. Definitely my daily driver.
3 Likes
3.7 is so good. Loving the experience. The agent mode feels noticeably faster and more accurate!
1 Like
Keeps failing and we end up paying like tons of requests.
3 Likes
3.7 is insane, really loving it. It does seem a bit unpredictable regarding what triggers consuming a fast request but this is nitpicking. It did a full refactor of 700 lines of code in one request, but then a bugfix with a few terminal commands cost like 3-4. This is definitely my new daily driver though!
Error connecting to anthropic. Please try again in a few moments. :(((
3 Likes
(see screenshot)