its a shame that such a revolutionary product is being gamified by costs. for $20/mo i get unlimited pretty-fast responses on chat gpt. this speed defeats the purpose of using the app, especially when a response is wrong
Why was this flagged and hidden?
how long you wait for slow request?
several minutes
Seems to have slowed by around 30x in the last 48h
Nobody likes to wait unnecessarily. However the constant posting of new âslowâ request issues without any details, without context, usage, session id etc. as the Cursor team all the time requests for easier checking on their side is just wasting all our time in the forum. The impression is that some people have not a good understanding of what they are doing with Cursor, but hey lets blame the company.
From your post its unclear if you lack understanding of how LLMs work and what happens where (at Cursor Edit, at Cursor server, at 3rd party AI server) or just wanna vent your frustration.
What is being gamified here?
You mix up a non-integrated standalone tool like ChatGPT with an coding editor with AI integration.
Why not use ChatGPT if cost is an issue? this is a real question.
Have you subscribed to OpenAI status emails? i see every day status emails from OpenAI that some part of their system is down.
Then you mix up wrong answers with slow answers. How often is nowadays the answer wrong? Have you tried better prompts? do you know how to write prompts for good responses? Do you use cursor rules? Any structure to prompting?
You can easily burn through 500 fast requests with bad prompts and just writing âcontinueâ or âyesâ on every question the AI asks.
Why not give more details and ask us others in the forum for feedback or suggestions? So many here are happy to help and share experiences.
One thing we do not need is one more slow requests rant post without details. Nobody says you cant express your opinion. but why not add it to any of the existing slow threads in the forum? at least there we can see who all has similar experiences.
We do not see people posting all the time how great Cursor is. Why? Because they are busy doing things with Cursor. Occasionally some people share their good experiences. But im sure there are more with great experience than issues, otherwise nobody would be using Cursor and it would not be so famous.
Unfortunately that fame is an issue. The more users join Cursor the more resources are used. 
Itâs been crazy slow for me. I thought I saw Cursor team say that the further you go over the limits, the slower things get.
Which is a fair way to do things!
That said, it would be nice to have this be more visible per request eg, youâre at 9000 over. That way we can pace ourselves a bit more.
Side effects of a great product: cursor makes things too enjoyable and we end up wanting to build too much
Agreed i would also wish for endless fast requests. Other tools i use do not even show the limit. Eventually you just get a message that you have to wait 3hours or until tomorrow to be able to use powerful models again which is more annoying when i am in a productive session.
Using different models for planning and the best for coding can help as well reduce your time in slow queue. Sometimes its also important to not just answer AIâs questions, but tell it what to do instead to avoid it going sideways.
Yes the more requests you push into slow queue the longer your own queue gets. This avoids slowing down users who do not push to often to hard.
Some who get good productivity with Cursor use on demand price after the fast requests, which allows you for a reasonable price to extend your run. Im frequently checking the status page to understand how fast i use requests overall. So yesterday i used 250+ fast requests on a project, but afterwards i recognized i should have not let the AI just continue in a circle trying to fix an issue where i could have guided it better. So today trying other models and keeping top models for coding only.
make it a better cursor team. your one of the fastest growing compnay & best .dont make it people unhappy, please. also possible to increase the 500 limit, please
Use your own API keys and see how much you will spend. Cursor. team charges 1/3 if not 1/5 of cost you would pay for your own API usage. Or you want them to pay for your API requests? I think they should just block the app after 10 slow requests so more people pay and they can invest more in the product. At the end, in a few years, we will have AGI, and you will ask the cursor to write you your own IDE that you can use for free.
@ yurkomik Actually, Cursor is generating significant revenue. By the end of 2024, they reached an annual recurring revenue of $100 million, growing 9,900% from $1 million in 2023. They have around 360,000 paying customers, including engineers from companies like OpenAI and Shopify. So, theyâre more than covering API costs and operating profitably.
Yeah interesting facts but how does that explain the OPs âgamified by costsâ, â$20/mo unlimited pretty-fast responses on chat gptâ or âdefeats the purpose of using the appâ?
Other programming IDEs cost way way more without an AI integration, and OP talks about chatgpt.
Whatâs wrong with ChatGPT? Itâs great at what it does, itâs just not integrated.
The problem with cursor is that without AI - itâs just another IDE. Thereâs no value that I can see if you stripped out the AI component - but cursor is selling a product, not a component.
The quality of the AI used bounces between superhuman braintube and complete idiot - and when it gets dumb as â â â â it start consuming my credits. It will forget access to the workspace, it will not realize the capabilities it has - it will edit files in the wrong project - literally putting python in c#.
Iâd pay fast credits for quality responses. I dont want to pay for the AI to turn into a vegetable - and this isnât something to blame on Anthropic. Cursorâs product is IDE+AI. If you position it as IDE w/out AI, no one will use it.
You have legions now of customers like me (technical product manager) who can build but donât need to fully code. If the product doesnât work for me, iâll go back to pasting files into Claude and just pay them directly.
I got the 20$ a week ago, and everything worked amazingly. Even after running out of fast requests, the slow requests were still reasonableâjust a few seconds each.
But over the past few days, the speed has become ridiculous. Itâs so slow that I have time to write frustrated replies in Cursorâs forum while it just sits there, stalled.
It really feels like a not-so-subtle way to push people into buying more fast requests. A shame, because itâs a great product.
Hope this gets fixed soon 
You can easily burn through 500 fast requests with bad pronmpts and just writing 'continue' or 'yes' on every question the Al asks.
How do I address this issue? It keeps asking and I need to keep answering yes
If an AI keeps asking questions, its because:
Overall:
So starting a new session(new composer/new chat) helps reducing the context overload and helps additionally to reduce mistakes as too much conflicting history in context may confuse it.
Would you mind giving examples what it asks you and in what situation? Also happy to jump to a separate thread about prompting if you like.
There is nothing wrong with ChatGPT of course but its impractical for productive work as its disconnected from where you work usually.
People work in apps (desktop, mobile, web). They write text in docs, work on tables in sheets, and so on. Having to copy paste from and to ChatGPT is just impractical.
This is where Cursor comes in and where other traditional IDEs have issues with AI integration. Lets take JetBrains IDEs. Their AI Assistant is not well integrated even they control the whole IDE. Cursor on other hand made the integration very practical. Yes other AI IDEs (AIDEs?) may have additional features. Of course Cursor without AI Integration is just VSCode but the AI is not on human level yet.
You got to stop it from getting dumb. Seriously.
Todays LLMs are so powerful and capable. Most code can be well writen by LLMs with good prompting and guidance along the process. This does require that you understand what you are doing and what may go wrong. It definitely helps to know the programming language or at least one programming lanugage.
I agree that some part of the integration need to be smoothed out.
However how would it edit files in the wrong project? Note that I open one project Cursors âWorkspaceâ. That avoids confusion.
You could also add rule files with file extension filter and mention programming language in description of the rule file to avoid mixing code between languages if you mix files in one project.
Im happy to share ideas that may help you if you like. Let me know what happens and how it goes wrong. That would also help others.
They werenât not very slow before. Until Cursor decided to making slow request to âreal slowâ mode. Like forcing it to do so.
Back when it just wait for few seconds.
Feel like they are forcing pro plan users to using their Usage Based Pricing to extened requests. (If they really did I think I just stick with Cline instead 
.)
Lets give them for now the benefit of doubt at least. there is no proof or at least i know of none that they are doing shady things.
But because its more busy it cant be exactly the same as months ago. Even if we wish.
At same time there are users who can handle their prompts good and avoid unnecessary additional spending on more fast requests or usage based pricing. It may not be always avoidable but we all should make an effort to do better.