It’s hard to hear such news, and I believe the team will fix it.
I had to use Claude3.5 yesterday with no solution, the code was a mess, and in the end, it was even worse than doing nothing for a day.
Unfortunately, intellicode/intellisense are not the solution, so I cannot select them as the solution.
I have a new idea. Anthropic has released Claude 4 and made its API available on Amazon Bedrock, so integrating the API might be a better option. I’ll purchase access later and test it out to see how it performs.
If you can tolerate Cursor’s slow request speeds, you could try rolling back to version 0.47.8. While it may be slow, it does work effectively.
Thanks, but I can’t tolerate such a slow speed either. I can only temporarily use the auto mode to get back to work (although the code quality is very low).
I had the same idea, so I bought Google Pro and tried to integrate the API to make the Agent work, but it gave me a limit exceeded message.
Claude 4, which was released yesterday, is also unavailable now.
I can only wait for the team to recover, which is very frustrating.
1 Like
same， can‘t use
I’ve also used up my 500 fast requests. The slow requests are almost unavailable. I waited for 10 minutes but got no response. Since I used up my fast requests, not a single slow request to Claude has been successful. I suspect that Cursor has detected that I’m using a student account and deliberately restricted my access. Are you all using student accounts too? If so, I’m willing to buy a new account, but I’m not sure if that’s the case.
1 Like
Similarly, I’m encountering the same issue. When specifying Claude 3.7 and Gemini 2.5 Pro, the system remains stuck in the “generating” state indefinitely, with no response returned thereafter. Conversely, Cursor only functions properly when GPT4.1 is specified. For reference, my Cursor version is 0.50.5, operating on Windows 10. It’s worth noting that this anomaly has persisted for approximately four days, emerging immediately after upgrading from version 0.49.6 to version 0.50. I’ve observed similar reports on the forum, where this problem has already been flagged by other users. Notably, rolling back to version 0.49.6 did not resolve the issue. Initially suspecting an IP-related problem, I tested with a VPN, but the issue persisted unchanged. It’s important to clarify that my account is a paid subscription acquired directly from the official website, ensuring full compliance with service protocols.
1 Like
I am using the student in the same way, first Claude 3.7 went completely from the menu, then I started using the 3.7 think model, and now the 3.7 think model is gone, and suddenly claude 4.0 came in the application (these happen while I am using the application) it is very scary that they intervene and update while they are using it at that moment, I perceive it as if the student accounts are eating limitations. so why did you open such a program if you could not afford it.
Same
GPT4.1’s performance is far worse than Claude 3.7 and Gemini 2.5 Pro
I can’t even use Claude 3.5 and the previous Gemini series
I have a paid team account with per-request charging enabled, which is very frustrating.
Mine is the same, not only the request is stuck in an interface and does not respond, but also the number of times is consumed. I have already wasted more than a dozen requests
Same Here, stuck at generating..
Cursor version 0.45.14, claude 3.5
I’m not a student account, I’m a pro account that pays $20 per month, same, can’t use
Yeah, I expereince the same. When select a premium model, cursor just stop or unresponse. meanwhile auto is working but generate bad code, wtf? Sending email to the dev, got no response.
the fact that no one has denied that this models are no longer usable for slow requests, just makes me think this is new business model. we are fkd
I encountered another situation: only when using the $0.04-per-request option does everything work without any restrictions, but once I turn off the $0.04-per-request option and rely solely on Cursor Pro, it stops working properly.
This can only mean that there’s nothing wrong with the third-party model itself — they’ve just changed to a different money-making business model.
(post deleted by author)
The developer clarified that the root cause of the slowdown fundamentally stems from their interception mechanism:
You have consumed all of your request allowance (usually 500 requests), and have used a significant volume of slow requests to the point where your requests are severely throttled
1 Like
Perhaps the reason is already clear: Pro users exhausted their 500 high-speed requests and resorted to using a large number of slow requests, which led to their responses being restricted. The restriction on Pro users’ slow requests is actually linked to the student discount offer, as the large influx of students using Cursor has increased its overall burden—and these two things are somewhat connected.