i still get slow request on all in composer,
not true, re-check and if its like that report in Bug Report
Hey, I went through your customer journey: did the trial, liked it, signed up paid for a year, and now your service quality has changed and the “solution” is to tell me to SPEND MORE MONEY to get the original level of service I paid for.
Who does this? How is that OK?
10 Likes
I would like to say firstly, that Cursor has enabled to do so much more than I ever thought possible without it. As an amateur, part time programmer - I’ve been able to complete a robust, real-time application with AI assistance. So, to any Cursor developers who are reading this - THANK YOU SO MUCH. You’ve enabled me to make something great! Maybe that’s why we’re all so frustrated right now because we’ve become reliant on this tool functioning in the way it has for months. This is a very sudden change in functionality, and, like others have mentioned - the current wait times make this product unusable. I might be willing to pay more, even pay per usage - but there has been little clarity on what my options are going forward. If this is an intentional change in the way the system works for paying subscribers - let us know so we can make an informed decision. Oh, and please, pretty please, with sugar on top… let me have my old Cursor and Sonnet back, and we can forget this whole thing ever happened (my preferred option).
9 Likes
Slow requests are just unusable right now.
Timeouts, connection failed, etc…
Will this be fixed or is it just the level of service we gotta expect for us lowly people who can’t spend multiple times the full monthly price, I blew threw my fasts requests in 2 days, mostly while not even using it.
So is our only option to pay more or just not use composer anymore?
Edit: Ive had a VERY simple test request run for 30 over minutes and it’s still on “Slow request, get fast access here”
2 Likes
Why not incorporating Deepseek v3 into Cursor which is much much cheaper than Sonnet in which way Cursor can save a lot of cost. I’d rather use deepseek instead of the current Sonnet with huge delay. We need diversification. Thanks.
13 Likes
I’ve just tried again, and it seems to be working now. I’m not sure if it’s only on my end, so could you test it as well?
Just tested, took around 15-20 seconds to start, which might sound like alot but it’s alot better than earlier.
The request did not hang or bug and it completed correctly.
I then asked a follow-up question, again it took around 15-20s but it also worked.
That’s a good improvement, I don’t mind waiting a bit if it works at least!
Hopefully it’s not just a temporary luck we have.
Not sure how accurate your estimate is given:
OpenAI Estimate:
Based on the pricing details you’ve shared, here’s the estimate for 500 additional requests using Claude Sonnet 3.5 with long context chat:
- Included Requests: You get 10 requests per day included in the Pro/Business plan.
- Additional Requests: After 10 included requests, each additional request costs 20 cents.
- 500 Requests Calculation:
- First 10 requests are free.
- The remaining 490 requests would be charged at 20 cents per request.So, the cost for the additional 490 requests would be:490 requests×0.20 USD=98 USD490 \text{ requests} \times 0.20 \text{ USD} = 98 \text{ USD}490 requests×0.20 USD=98 USD
Thus, the total cost for 500 requests would be $98 USD.
So in a nutshell it looks more like this to me:
100 bucks for 500 requests according to table I found on the account in Cursor AI. If I need 3,000 requests a month, that’s $600/month… that’s chunk of change or am I getting it wrong?
Codeium gives you this for 1/10 or $60/month:
Cascade credits:
-
Infinite premium model User Prompt credits
-
3,000 premium model Flow Action credits
-
Can purchase more premium model
credits →
$10 for 400 additional credits with monthly rollover
- Priority unlimited access to Cascade Base Model
1 Like
@eraysahin06 @webdo
if you’re only seeing 20 sec waits rn, just give it time. the more you use it, the slower it gets (that’s literally what the post said). i just tested it and it’s already taking like 3 mins for me, and i bet it’ll hit 4 mins pretty soon.
I’ve been using it all day, as you can see on my previous reply, I was at over 30 minutes of wait earlier so don’t worry, we are all in the same boat as you. Slow Pool Information - #29 by webdo
Is there no solution? Claude, I had the coding done in 5 hours with Cursor, while it normally takes me 1 hour. Will there be no solution? Will Cursor always work this way?
My estimates were for Claude direct API usage, I wrongly assumed you wanted that and not Cursor costs.
Nope, just go to Settings | Cursor - The AI Code Editor and click ‘add fast requests’, you will see it costs 20$ each 500 added requests, usage-based pricing is for emergency only or should be used as such.
1 Like
Ok sorry to keep going on this but it’s a bit confusing.
So there are really 3 ways to pay for 500 additional large context Claude Sonnet 3.5 requests:
- Settings | Cursor - The AI Code Editor → add fast requests at $20/500 requests
- Settings | Cursor - The AI Code Editor → enable usage-based pricing at $100/500 requests ($0.2 per request)
- Anthropic API Key (in IDE settings) → your estimate $15.00/500 requests
Thanks for clarifying!
3 Likes
Still extreme slow for me. Now I have to rely on free Ai: gemini-exp-1206 in Cursor. And Gpt4 is also slow as Sonnet. So it is not Sonnet problem?
Just came here after waiting 1-3 minutes for each message. Glad to hear I’m not the only one experiencing this issue.
Though in 2023, I remember reading on the Discord that adding models other than GPT 4 was very difficult, but it seems you’ve solved that issue considering the models you added recently. At this point, I don’t see the point in not adding Ollama or OpenRouter.
These are great fallbacks when stuff like this happens.
But the key point here is that without the “Agent” mode, Cursor is no different from competition.
I get what you’re saying, and there’s some truth in it, but Cursor stands out for being as complete as it is. Every “what if” in Cursor is solved, and it makes the user experience 10x better.
The fact that you can undo applied code (CTRL + Z works flawlessly with it) is already big because a lot of other editors don’t have this. As someone who’s building the future of auth, handling every possible edge cases, I’m impressed by what they have done.
1 Like
A lot of people who try the trial end up subscribing. If they didn’t have this, I might’ve not subscribed for almost a year, month by month. Because it would’ve been very hard to make me pay.
After trying the trail, I was convinced I needed it. They make more money from that, so removing it would hurt their business even more.
Though I can also see your perspective: they have a lot of paying users now, so they got the social proof they need. If Cursor was talked this much about a year ago, I would’ve probably purchased anyways.
But trying out a product before full commitment is still important.
literally unusable now crazy how you can’t even use gpt4o without a long delay
The price list seems outdated because long context got removed?
Would be good to know if composer agent usage is under “Fast premium models: As many fast premium requests as are included in your plan, 4 cents per request after that.”
That would help people who are >500 and <1000 in usage and want to save some money. $20 / 500 = $0.04.