Looks like it’s time to unleash the Deepseek . Not only will it ease reliance on Anthropic models it also provides better options. I doubt anyone uses gpt-4o lol
3 Likes
Honest question. I have work to finish.
Can i pay my way out of this problem?
Does business subscription solves it or any other pricing add-on?
1 Like
We are without a doubt their largest customer.
3 Likes
Well time you asked for your money back then 
2 Likes
Same here, i hope it’s just a temporary bug, because “unlimited” should mean “unlimited” not “enable usage-based pricing to be unlimited”, i’m sure that’s just a temp fail…
It is suggested to solve the problem of unlimited spam mailboxes. A large number of such users are destroying them. I think this is one of the reasons.
It may not be ideal but it seems that I’m able to continue working by switching to the haiku model of Claude.
okay so i’m not the only one, gpt4o is just not the same -.-
GTP4-o is an a-hole 
But seriously, any ETA on a fix? A permanent one?
1 Like
Sonnet is alive.
3 Likes
Business plan has the same limits as pro, so that won’t help.
In theory you can pay your way out of the problem by using an API key and paying for every token yourself, that’s of course IF Anthropic or Amazon Bedrock APIs have enough capacity available.
1 Like
Yeah, it’s back!!
2 Likes
I wonder if Snyk is doing a Dodd via consuming a f ton of free accounts and spammin cursor apis for prompt transfer infra overload?
Would be great to eat insight into how cursor angles such loads/attacks…
Checkout the thread on HN
Deepseek is on other side on planet (CN). If they wanted they would added it 2 weeks ago. For now they just switch slow requests to 4o or Haiku
I have this problem too, ■■■■ it, I was thinking my pro period is over.
Yes, that scared me because it is unlimited.
But the problem is solved by now. Its working again.
1 Like
Generally speaking, this has to do with Cursor’s business model.
They buy tokens in bulk and then aggressively manage their own context in order to avoid high-token conversation arrangements in Claude. Think of it as being something like
- Send the last 7-13 conversation segments
- Send an internal cursorsmall summary document, likely xml or md, to hold the illusion of long context together
- Send a summary of the code the user is working on along with specific samples; likely occurs in an agentic context
This allows Cursor to be highly effective while also rationing tokens and making the profit required to stay in business without charging everyone 3-5x the cost they expect to pay.
Slow requests are, in essence, what they allocate in terms of free tokens as a loss leader to get people to keep using the tool between purchases. They’re a necessary evil that cost Cursor money but without them people would exit for other “free” tools with basic AI tiers. In other words, if you give people just enough to want more they’ll buy back into the fast lane and your company makes a profit.
What you’re seeing is the popularity of Cursor: there aren’t enough slow tokens at peak hours to satisfy all of the free or low-cost subs combined with the folks who are trying to white-knuckle the last few days of their expired sub without paying another $20.
3 Likes
They should foucs on sonnet 3.5 all other ■■■■
It’s possible to use local computing via Ollama or something?
1 Like