As we all know, Cursor’s slow premium requests have been slowing down to an unbearable level lately. Those who have purchased a monthly or annual Pro subscription are now regretting it. When will Cursor’s developers literally inform their members? Will there be a new subscription pricing? Or will the free trials end? There are so many questions waiting for answers but no official statement has been made yet. Yes, I know @Amanrs has a Slow Pool Information thread but this doesn’t show any validity because it says if you want speed, spend more money. When will there be an official statement for a truly permanent solution?
Hey, we don’t have any further to announce above and beyond the slow pool information post from Aman.
The post is intended to explain why queues may be high with the slow pool for Anthropic queries, as they are unable to provide us with all the resources we need and, as such, we prioritize fast requests, which in turn causes a large queue on the slow pool.
The slow pool is always available, and is unlimited, but is intended as a backstop for the last few days of the month if you run out. We offered it to ensure you weren’t left with only non-premium models before your usage reset.
The permanent solution right now is to either:
- purchase more fast requests
- wait in the queue for the slow requests
- use an OpenAI model, like GPT-4o, as the queues may be lower
If you’d be willing to pay more to not have to queue (as suggested when you said “Will there be a new subscription pricing?”), then you can enable usage-based pricing in the dashboard, and get charged $0.04 for every fast Claude 3.5 Sonnet request you make outside of your included 500 requests.
Many users, especially those on a budget, are happy to wait in they queue to avoid paying more, but we give you the option if you choose to do so.
one day you will reveal when “end of the month” is
because on the usage page it only shows last 30 days
I’ve never heard of a service that you PAY for, that penalizes you the more you use it.
Taken from their official statement that the more you use, the longer you wait.
Anyone else think that’s OK?
1 Like
Nobody likes unnecessary slowdowns. However its also not entirely fair to claim that Cursor penalizes you the more you use it.
Since LLMs are not free and most apps incl. ChatGPT or similar just tell you that you have run out of your limit and have to come back hours later, Cursor seems to put the requests in a queue.
Its also logical if you launch many slow requests into the queue, you will have to wait, the more you queue the more you wait.
It would be completely unfair to other users if one user can add many slow queries into queue and next users slow request would be set after that user instead of after that users first slow request. This seems to be more fair for all users and penalizes abuse.
(I am not affiliated with Cursor, just my opinion)
1 Like
Hey, you can check your refresh day by entering the billing section on the Cursor website. The day your monthly invoice is taken is the same day your usage will reset.
As I said in my original post, the slow pool is meant to be a stopgap, in case you run out of fast requests before your renewal date. Therefore, we are aiming to allow for minimal issues for those who use it as such!
As an example, to be clear:
User A:
- Pro User: 500 fast requests ($20/m)
- Used 520 in a month (so 0.04% of allowance over)
- As slow pool usage is low, their queue time is low
User B:
- Pro User: 2000 fast requests ($80/m)
- Uses 2100 in a month (so 0.05% of allowance)
- Relative to their plan, their overage is still low, so their queue time is low
User C:
- Pro User: 500 fast requests ($20/m)
- Uses 3000 requests (6x their allowance)
- High slow pool usage, means high queues
Obviously, in this scenario, our priority is to the customers who use the “free” allowance the least, as it is there as a backup. User A and B both used it as a backup and therefore have the least amount of time waiting. This allows us to keep requests fast, which not reduce the allowances or increase the base plan price.
For the best experience, User C could’ve paid for the additional requests to avoid the queues. However, for whatever reason, they choose not to - which is fine. However, as they increasingly use the slow pool (and therefore adding to its usage and reducing its capacity for other users) they will be increasingly penalized to incentivize them to pay for the requests, or to use other models. They can continue to use the slow pool, for free, with no usage limits, under this restriction of speed.
At any time, all of these users can still use the slower models, which for >50% of queries would do an adequate job. Also, we have now separated the slow pool for OpenAI and Anthropic models, so switching to OpenAI models would provide the user with a shorter queue on the slow pool (as it is Anthropic we have capacity issues with).
We’d love to remove the queue and give unlimited fast requests, but we make these decisions to ensure our company and pricing remain stable and viable for the long term! Any decision one way or another can cause a domino effect, so we are very careful and thoughtful about these decisions.
I hope this helps you understand why things are the way we are, but we are always working to improve the product and the pricing around it, giving you as much usage and features as we can for our plan.
so how do I check this month’s usage ?
I had no idea I was coming up to the limit because it only shows the last 30 days
Hey, logging into your dashboard, I feel like the number shown there should be for the month, not the last 30 days (despite what it says!).
Let me confirm, but either way, I can see this being confusing and, if it is the last 30 days, not a very useful statistic to show.