Same here, lost a lot of request this morning. Switch back to Copilot for Github for the morning.
same here
Even when it fails, it keeps debiting premium requests. Also, it doesn’t even say clearly on the billing page that it does count towards premium requests.
This is quite likely not even legal in the European Union.
I just realized (after using it a couple of times today, not for long) that it has already consumed a large amount of my premium requests this month.
I used to get the errors that the others here are citing. But now it just gets stuck at “Generating”. I keep retrying but I’m wondering if that’s just making it worse for your backends 
Do we have a sense of when Anthropic will have enough capacity with this increased traffic?
Really looking forward to it
So far it either couldn’t find the files I added to the chat + referenced with @, or had an error connecting.
At the moment I’m using it with Claude Desktop since I’m only editing single files until the kinks are ironed out.
My impression is Grok 3 (planning), and Claude 3.7 (implementation) will be the dream team.
In other news:
-
OAI employees are dropping cryptic posts on X again, and since they also recently released SWE-Lancer I wouldn’t be surprised to see 4.5 this/next week and be a step up from their other models for coding (low bar!
) -
DeepSeek is also pushing up the release of r2, and Google’s Gemini Code Assist now has a generous free-tier (search for it in Extensions.)
Fun times ahead!
my project is so big that claude 3.5 sonnet is hallucinating… 3.7 was hallucinating a little less and was actually helping me… but unfortunately, it’s offline. any idea when it will be back in slow mode?
I’ve been hammering on 3.7 hard since it came out and here is what I’m finding: response time is not good on a larger existing (Electron) project w/ and w/o thinking. However, on a brand new (Rust) project, using -thinking only, it is AWESOME and more specifically - much better than 3.5 in terms of code quality and agent focus. I really like how it is slowing down (a reasonable amount of time) before starting to code. I was getting really tired of the 3.5 teenagers refactoring my api service when I only wanted them to focus on the hello world UI ;-). I have a lot of faith in the Cursor team based on many hours on 3.5 that the perf issues w/ 3.7 on larger codebases will get better. So for now, I’m using 3.7 on newer projects and sticking with the 3.5 kids for existing. Keep up the great work, you’ve built an awesome product!
Hopefully, Cursor team would gain a better discount from Antrophic, compared with teh api price of retail version.
I suspect this is due to the auto-reasoning feature even for the non-thinkg 3.7 version.
Available my *** !
We’re experiencing high demand for Claude 3.7 Sonnet right now. Please try again in a few minutes
Since last night 10 hours and counting.
Just remind me again why I pay for the Cursor?. I can watch some traffic light blinking for less than 20USD for sure.
I get this message since upgrading to 0.46.4, while the premium request counter keeps increasing.
How can I join their beta version?
claude-3.7-sonnet extra_headers={“anthropic-beta”: “token-efficient-tools-2025-02-19”}
What if I am using my own anthropic key. is there a way to enable 3.7, I have pro account with cursor and a team of 20+ developers on my account we all use our own api keys exclusively and still I d not see a way to enable 3.7 please advise
On cursor, does claude 3.7 sonnet thinking version think before taking any action? Or does it think after each query from the user?
Apart from a few queries on the first day of the release which aborted half-way, thus rapidly consuming my credits, I have been unable to use Claude 3.7. What exactly is it that I am paying 20 USD for, if not to have this available? The first use was a blast, it really got me hooked. You are onto something cool here. Now I have some deliveries where I would love to use Cursor but it is so extremely unreliable that I cannot confidently use it in my work
Edit: wait a minute. I enabled usage based pricing and now it works. I suspect the buggy release ate up most of my 500 queries. Any chance for getting some extra queries there?
3.7 is running very slowly, and most of the time, the system doesn’t respond. I keep asking it to retry, but I have to do this multiple times, and it still doesn’t respond. I’m wasting my credits for nothing.