Ya, that’s what I’m noticing. I’ve been wondering if it’s just lagging or is it really not counting.
Yeah I have paid and all of my premium requests are timing out. So I’m in the position of paying and not getting.
I’m not sure how the additional requests pricing works. It has cost me $93 for additional 500 requests per month.
1 Like
- Stop the free account: There are even software that allow to change your machine ID and you can basically get the 15 days trial over and over. I think a free trial is not the good approach, anyone who want to test should be able to pay $20 to give it a try. I think by now Cursor is popular enough that people know what they are signup for.
- The composer agent use so much token, maybe the agent should be available only with fast request. The slow pool should not be available for that.
- Stop the abuse: There are people abusing the free account, but there is also people sharing 1 account with 8 persons. There is even an online community that facilitate that. That certainly put extra load on the slow pool.
- Introduce more alternative model. Haiku should maybe count toward the unlimited model to incite more people to use it. Deepseek is a good addition.
- Update VS Code, so we can finally use GH Copilot Sonnet 3.5 for chat and move some of the load away.
4 Likes
wtf
in your product order said No Limit but now you said
Wait times are calculated proportional to how many slow requests you’ve used so far, so wait times shouldn’t be bad unless well over your fast request limit.
wtf
So why should I pay for a subscription, $20 in Thailand is not a small amount. If the solution is to pay more, wouldn’t it be better for me to just use ChatGPT directly? Why should I pay you, if the more I use, the slower it gets? Isn’t it better for me to use the free version with 500 requests per month?
2 Likes
I agree with all points except number 2. The idea that the slow pool shouldn’t be available for everything is, naaaa. It’s called the slow pool for a reason, and it should be accessible for all services I’m paying for. The problem isn’t the use of tokens; it’s that the system hasn’t scaled to meet demand IMHO. The lack of scaling cpuld be due to the vendor providing the inference not being able to scale too. The slow pool is a good concept because it doesn’t guarantee immediate availability. It acts as a kind of failover, especially considering different time zones and varying coding times, meaning the pool will naturally fluctuate in usage. If there were enough resources to meet demand, this wouldn’t be an issue. The real question is: what wait times are people willing to accept for slow pool access? Should we onboard more users without scaling? What alternative vendor options exist to handle the slow pool? In my opinion, managing expectations is key. The app should clearly display the number of users in the slow pool and expected wait times. We should also cap agentic iterations on the slow pool, for example, 500 iterations within 4 hours for paying users, after which users would experience longer wait times. However, users shouldn’t be penalized for using the service, as this would undermine the business model. Instead, a fair-use policy should be implemented with transparent performance metrics for the slow pool. Free users should have reduced access, but not be completely cut off, perhaps with 30 agentic requests per 4 hours. The core issue is transparency. Without it, we’ll all assume we’re being treated unfairly, when in reality, the team’s hands are tied due to resource limitations beyond their “control”.
Ideally we had prefer a cap on number of users until scaling demands are met. However, due to their popularity this is probably unrealistic. I also have a slight suspicion that not everyone uses the software for coding which could also bring in the so called unwanted users.
Yo, don’t give up on that dream bro. There are many alternatives like GitHub copilot which is being offered for free now, Gemini which has gotten good and free, there’s Continue.dev which can use local models and free/cheap ones like DeepSeek V3.
Now is the time, because every company is offering AI for free. I believe in you, and I hope for the best man!
1 Like
Why does composer keep prompting me to ‘get fast access’ when I already am paying for it?
Hello,
Please add a short cut to switch between two models: I want to switch from 3.5 sonnet to 3.5 haiku or gpt-3.5 because not all requests require me to use a fast model and that wastes my fast request pool.
Thanks
Forward slash key should allow you to cycle through all of your enabled models.
Here’s my two cents on this topic.
I’ve always had serious doubts about whether cloud-based AI solutions (not limited to cursor) are in any way profitable; especially those who build on top of other AI service vendors (OpenAI, Anthropic, etc.). Companies like cursor also have employees (assuming it’s not all AI that build their products XD) who don’t pay themselves: someone has to cover for their costs. At an early phase this can be investor money but that’s obviously not sustainable so at some point the company has to either refine their product to such a cost effective level that they can keep their initial pricing or gradually (in this case quite drastically) hike up their prices to compensate. I’m not trying to point fingers at anyone but rather this is just true in general regardless of business (as you can see in other areas too like Uber Eats). It’s difficult for brokers to grow when their operation margins are so slim.
Is cursor valuable to a point that I am willing to pay a fixed monthly fee? Yes. Is it worth 20 USD? To me, yes. Do I think it’s underpriced? Logically yes (considering the cost for labor and infrastructure), but personally no. So what am I going to do? I think I’ll just live with the completions since that’s already powerful enough and do the thinking myself. At the end of the day that’s what I’m being payed for at my company so I don’t really have any issues there.
It would be wonderful if one day we can all be liberated from “writing code” and focus on “building products”, but sadly the reality seems to be we’re still not there yet. But until then I’ll invest my 20 USD per month on the intelligent completions and stop touching the composer as often so Cursor can rake in the profits.
1 Like
This explanation is contradictory in nature . If thee were truly a bandwidth constraint due to an unexpected usage increase, nothing the user could do would effect that fact. From the way this post reads, the bandwidth and capacity exist and are being withheld unless, more money is spent. I also haven’t seen a fall off regarding Anthropic’s response time in any of the myriad of applications I use to interface with Claude. Even Cline is still slugging right along. So I’m not entirely sure this is a fully truthful statement. However, you guys have got to keep one foot in the world the rest of us live in. We don’t have hundreds of millions of dollars in investor liquidity pilling up around the office. Some of us have clients who are already invoiced and already paid who gave us so much grief about the initial invoice cost that going back to attempt to upsell the invoice to cover the costs of buy back functionality we already had, will only end with “well if AI can do it what am i paying you for in the first place”. You guys have put yourselves in a position of being integral to people’s livelihoods. People depend on you and your platform to feed their kids and keep the lights on. And solutions that end with “give us more of your money” feel disingenuous at best and crooked at worst. It’s been a couple weeks now, maybe you guys need to work a deal with Anthropic to stand up your own datacenter for your users exclusively. Musk did that in 90 days. I love courser, to the point I’ve put many eggs in your basket. Please don’t make me eat omelets for dinner. I hate breakfast for dinner.
8 Likes
They’ve also just got a lot more funding now ($105m), and oddly it seems that the more funding they get the worse the product has become, not very promising.
You would expect it to be the other way around
Can you please explain more. How exactly does that work? Thanks
So your suggestion is that instead of charging 4cents for each extra request, the company should… spend $100+ million building a datacenter? To solve a temporary capacity issue that can be fixed by either:
- Using Haiku
- Using other available models (free)
- Paying 4cents per extra request
Im not sure if this is satire or real
1 Like
This information really needs to be included here: Get Started / Usage – Cursor
This does not convey the information:
users who run out of fast premium credits will be moved to a slow pool, which essentially is a queue of users waiting for a fast premium request to become available.
I’m not going to argue, but these recent changes made me check out the competition. This is the first time I’m doing this. I’m sure there are more users like me. Your bottom line for this month will tell you more, and will ultimately cause you to compromise or stay the course. Ultimately, we all vote with money.
3 Likes
lol , i think people got so used to typing an unoptimized prompt then start screaming about usage being utilised too quickly
1 Like
From Cursor’s perspective, the 500 fast calls are the primary offering, and the unlimited slow calls are meant as a stop-gap—a perk rather than a core feature. This approach is likely seen as generous, ensuring users aren’t completely cut off after exhausting their fast calls.
However, from a user’s perspective, that distinction isn’t obvious. The way the subscription is presented—“500 fast, unlimited slow”—implies that unlimited calls are a fundamental part of the service, not an extra or a temporary fallback. Naturally, users interpret their subscription as providing unlimited usage, with the only tradeoff being speed.
This difference in expectation leads to frustration when users discover that the slow pool is so slow that it’s practically unusable.
If the slow calls are intended as a limited fallback rather than a reliable feature, it might help to clarify this more explicitly in the subscription details to better align expectations.
Additionally, it’s not clear at all that there’s an algorithm in the backend that adjusts wait times based on usage, effectively penalizing frequent users. If someone uses the service as marketed—believing they have “unlimited slow calls”—they may end up in a system that deprioritizes them over time, which could be perceived as unfair.
Since this isn’t mentioned anywhere in the marketing materials, it creates a disconnect between what users think they’re paying for and how the system actually works.
1 Like