Make a new tariff plan with unlimited Fast requests or normal working slow of Claude 4.
If there was a cost of $40-50 for such a plan, I think many coders would be willing to pay. Right now, slow queries don’t work at all. They don’t make sense.
Oh, sorry I misread.
I agree that “slow requests” don’t work those days. Unlimited slow requests was the reason being why I paid for a subscription. But then it was supposed to be slow in exchange for helping the AI providers better manage the load.
Now it seems like I cannot even have a more than a dozen interactions every hour,
[EDIT: to be precise, I tested the time and it took 4 minutes for the start of the agent response with gemini 2.5 pro. Almost 5 minutes in total. On the bright side, there was NO noticable latency between agent interaction. 5 minutes to get my completion is a theorical maximum of 12 requests per hour, and that is only if I had already prepared the next request, and keep looking at the screen for the end of the completion, so in practice it is going to be even less, so I was actually nice saying no more than a dozen interaction per hour.]
It’s hard to call it unlimited anymore. I would actually prefer they directly tell me I am limited to X requests per day and have those passed in a reasonable time.
So, I propose something different, I propose unlimited REASONABLY FAST requests, just fast enough for the AI provider to manage the load on the servers and provide discounted pricing, but not have to wait just to make sure you won’t pass too many requests and cost too much per day.
[EDIT: but wait, they don’t call this a request, they actually count that as multiple requests. Which is kind of fair, except for the fact that I had to cut all my files and use the agent, because the 1M token context window was not respected and my attached documents seems to be cut after 10000 tokens… but then that might be because of using the previous cursor version…] Oh gosh, so many things to take into account… will keep trying with the newest version and hopefully things will become bearable once again.
Gotta be honest, if I had this, I’d pay $100/month and I’d easily use that in 5-cent chained tool calls to do guided refactors within an hour, guaranteed. Unlimited Sonnet would do it for me. Opus would be gravy. But I’d chew $100 in no time.
Alas, if Cursor has to pay the model, they need to pass that cost along.
Until I have a Claude 4 Opus in my basement…
Slow requests were temporarily disabled by Cursor team for Claude 4 due to huge demand for Claude 4 and limited availablity from Anthropic. (everyone is trying it)
Unlimited fast requests are not realistic, even though I would like them.
As long as AI providers charge for it they arent free for Cursor.
Ideally a higher package with some volume discount would be great.
I am paying now for fast requests and even then I get rate limited on Sonnet 4. Reality is Cursor has no Moat, is totally dependent on Anthropic and now I am considering anthropic max with Claude Code in VSCode fork
Is that what you’re talking about
We’ve hit a rate limit with vertex. Please switch to the ‘auto-select’ model, another model, or try again in a few moments.
Even if I pay, it doesn’t give me work
Hey, we’ve been hitting our limits here, but should be better now, apologies!
100% people don’t like to watch the API cost and think about how much money they may waste on a request that does not come through properly. It sets you up for constant psychological pain. Coding is painful already. People just want to use the AI without thinking about cost.
I dream of the day I can pay $10 a month and not worry about rate limits or cost and not have to track my number of requests because we finally hit a point where Coding LLM’s have become cheap enough, efficient enough & powerful enough that the average user can do all their coding tasks everyday without thinking “Oh I need to top up or oh I’m about to hit my request limit or now we have a slow queue etc.”
I don’t think we’ve reached that point, or even got close to it. Closer? Definitely.
On a big codebase, it’s not powerful and efficient enough even with unlimited budget, due to context issues and tools errors. Same with less common stacks or combinations of libraries.
Today, we’ve got it with Gemini 2.5 Flash, which I’ll assume is on the level of top tier 2023 models but cheaper. I’ll assume by 2026 we’ll have new cheap models at the level of let’s say Claude 3.7.
But is even Claude 4 Opus enough for all coding tasks? ■■■■ nope. The censored opposite of Heaven nope.
GitHub Copilot+ now costs $39 with 1500 credits. It would be great if Cursor did the same. I believe 1500 should be enough for normal use, and it seems like a win-win solution for both.
Unless they offer an ALL-IN plan with fully featured frontier models (no throttling, watering down etc.) Cursor will be marginalized. I just cancelled my plan to get Claude Code max plan.
Seems like OpenAI, Anthropic and xAI are very interested in providing codign products themselves. At some point, Cursor et al. will simply be cut out of the deal. They don’t control their input factors and are very vulnerable. They have very little secret sauce..
There is a place for a tool such as cursor ai, that let user test and choose many models directly in their IDE. But for sure, the quality of the service must improve with time, not degrade with time. That the service is getting slower than before, now consistently requiring you to wait several minutes every single time, for usage of a model otherwise free to use online, and to the point that I have to complain every day because it’s still not solved, is not acceptable.
Slow requests were temporarily disabled
They are still very much disabled…
There is no such thing as unlimited. LOL
Hello everyone!
I think we really need a more expensive plan that takes into account unlimited access without the max mode. I’d rather pay 50$ a month and have no access problems. Currently we still can’t use Claude 4 in slow request. Of course, we need to increase the 500 requests, which now, with all the thinking of an AI, take up to 4 days.
I hope the cursor team will take this into consideration. A higher package and why not include a discount on max mode.
Because now there are too many abuses of multiple account sharing. It is difficult to support unlimited normal user experience. Now Claude 3.7 is slow for 20 minutes, which is unusable.
There really is a way to offer a higher package with some extra features:
- 200 fast requests per day
Or - slow request for max unlimited mode
Or other ideas, potentially.
Because right now, pro mode is pretty complicated. I can pay 50/60$ for a better account. Please team do something!
Thx
Something like this would be ideal for those that need it
Likewise, maybe even a lower tier since there are often requests from those in less wealthy countries e.g. $10 (tab, 200 requests), $20 (current), and $40/50 (more)
Or even just “add-ons”, such as a flat amount of requests per month for a price