I am already a pro user. I used 3.7sonnet two days ago and the effect was very good. However, I kept encountering errors when using it yesterday. I checked the official website and found out that you have cancelled users with slow requests. Then what’s the use of the money I recharged? 500 fast requests a month are not enough. I can accept slow requests, but now users with slow requests are not allowed to use 3.7sonnet. I don’t understand. When can this restriction be cancelled?
4 Likes
Currently, these limits are set by Anthropic, not by Cursor. The Cursor team has a maximum number of queries per second that Anthropic will allow.
They are in talks with Anthropic to raise these limits. However, Anthropic servers are currently under very high demand.
This is not an artificial restriction imposed by Cursor. It is due to sheer demand on Anthropic servers. Hence, Cursor is triaging premium requests to slow requests, such seems like the most reasonable thing to do.
Once Anthropic has more compute available and raises Cursor’s limit, they will be able to raise these limits on slow requests.
See here for more details: Claude 3.7 + GPT 4.5 - now available in Cursor!
3 Likes
In Win… free for free users… and a free trial, what is this limit in Cursor? This is not professional, and what is happening is not acceptable. In Trae also free. Too bad, Cursor, you disappointed me.
I hope this problem can be solved quickly. My pro version is about to expire. I hope I can continue to use 3.7 before it expires. If it still doesn’t work, I will no longer use the pro version next month, or I will use other software. I am a little disappointed.
This is the most wrong approach to take right now. Everyone lines up and if it’s slow, it’s slow for everyone. Why are we paying $20, what’s the point now?
Sorry but this is not answered. Cursor as a business can not operate if their supplier just bricks all models and still charges for access that doesnt work. It does NOT work even with usage based pricing enabled.
Why are we paying $20, what’s the point now?
$20 grants you access to 500 premium requests.
Right now, in addition to that, you still get access to slow requests. Just not with the latest models, which have only been released in the last 72 hours, and are only being restricted due to limits imposed at the provider level.
Cursor as a business can not operate if their supplier just bricks all models and still charges for access that doesnt work. It does NOT work even with usage based pricing enabled.
Cursor operates with multiple suppliers and many models, including:
- Proprietary models such as Anthropic and OpenAI
- Open source models hosted on their own compute, such as DeepSeek models
Cursor does not “charges for access that doesn’t work.”. This is a disproportionate exaggeration of the current situation.
What’s currently happening is that free users cannot access the latest model that was released in the last 72 hours due to restrictions at the model provider level.
You still have access to many other capable models from other providers, as well as the same models you had access to last week, while this temporary issue is resolved.
Please see Claude 3.7 + GPT 4.5 - now available in Cursor! for updates.
Hi Jack, yes i understand that Cursor depends on suppliers. That is what i mean that there is an issue the way Cursor ↔ Anthropic situation in last few days works.
I am not exaggerating at all. While i have much less issues than majority of forum users, it got also for me to a level where the setup is unusable compared to few days ago. And that says a lot.
While we all understand its temporary, its also something that is not the first time to happen and it will likely happen again the more popular one provider becomes.
Although yes i can use other models, the prompts, templates, handling, rules, docs etc. is optimized for a certain procedure that worked so well until few days ago.
And still i urge Cursor to explore with Antropic how to handle this in future better.
I am not expecting things for free or immediately, but Cursor surely must see how much this situation deteriorated, so its prudent to be prepared.
With ‘all models’ i meant Anthropic models claude 3.5 and 3.7.
3.5 is a well known and was until few days ago one of the best ones for most cases.
We do not have the time to play around with all models, time has to be used productively and appropriately to the cost and income. That was in last few days in very limited time frames usable (night time, vs regular work time).
The issue with 3.5 was not just from prompt perspective or results, but also lots and lots of issues with Cursor internal prompts polluting chat output, confusing models, UI issues, coniciding Cursor IDE upgrades, but also other things like YOLO not working, perhaps an issue with documentation details for such use cases if it needs specific handling apart of turning it on. etc.
Time is flying and there are not always times we can just wait until things ‘resolve’ eventully.
Please note that my comments here are not based just on slow requests but even worse its happening on usage based pricing, which is where it hurts more as productivity dropped significantly with increased requests needed to stop models from going haywire.
It does not take into account the stress, impact on mental health, etc.
1 Like
Besides my other reply just above this one where i forgot to mention you.
The lack of reliability is the main reason we have not comitted to an annual plan. (again not expecting anything free or special treatment, just a tool that does not negatively impact productivity of teams).
Thank you for your feedback. Rest assured these issues are being resolved as fast as possible 
1 Like