Hello, I purchased the annual subscription to Cursor not because it was the best editor, but because it genuinely offered 500 requests that would reset periodically even if unused. That was both an advantage and a limitation—what do I mean? If I used up my requests, I could pay for extra ones, but if I didn’t use them, they wouldn’t accumulate. It was a reasonable balance.
But now it turns out that, on average, I was regularly using only around 200 requests. With this new plan, to which I was moved without my consent, I was downgraded to a “freemium” model after just 50 requests. This version only allows for automatic completions, which often results in unusable or messy code that I end up discarding and rewriting from scratch. I originally chose this tool to help speed up my work—not to make it more difficult.
To make things worse, I sent an email to ask how this will be handled and whether there’s a way to return to the previous system, but I haven’t received a reply in four days. Is there a way to request a refund or revert to my original subscription? It’s absurd to move paying users to a new payment model they never agreed to.
To give Cursor some slack - it was clear it will not be easy (read: possible) to continue what they were doing and not be screwed up by provider’s prices. Which doesn’t mean they shouldn’t communicate way better, hire PR, consult CS experts, etc, what adult companies normally do in such cases. And being hurt after being told that you broke the contract(s) and people are calling you out on that - well, that’s beside my understanding really.
I used 20$ one, I switched later to 60$, for the sake of the test. As I said before, even 200$ is not a problem, if I know what I am getting for it, but with 60, and using it very very carefully and very rarely, I already got the warning I am reaching the limit. I’m trying to be reasonable here, and will just sum up what I mentioned in multiple threads:
Cursor rules are not followed anymore. For what was one prompt before, now I need 3-4 at least.
Sonnet 4 (and 3.7 as well) became overly eager to do the things I very specifically said not to do, and not to do things what I asked it to. Counts as zillion tokens and prompts.
In last couple of weeks I can not count anymore how many times I wrote “Are you stuck?” and “You are stuck, please continue”. Yeah, each one is another prompt.
Cursor is rewriting files for no reasons (or agent, whatever). 60 lines of simple TSX file suddenly gets written, then searched, then read again, to conclude it already exists, even it was written like a minute ago.
Rewriting quite small files when it decides ‘let me rewrite this’ for, from my point of view, no reason. Also rewriting big.
When pointing to what I wrote above, ‘prompting skills’ are pulled out, forgetting that some of us are, with all due respect - software engineers, and quite skilled to decide what code makes sense what not.
And worst of all - all those issues mentioned are directly impacting amount of tokens and price. I have no doubt (I really don’t) that Cursor devs are trying to build the best product every, and that they are putting shady prompts into it, but I have big doubt if they forgot to hire QA when they forgot to hire PR.
Saying this - testing Claude Code. Even with the cheapest subscription doesn’t have the issues mentioned above. So it is not the model. And will make my decision soon, as Dan requested in that other thread.
My Feedback is simple: I’m now pretty much priced out of using Cursor. Automode is bad and I want my slow mode back. I’m not writing code anymore, because every prompt I type I worry my limit is reached. This results in me not doing anything because it takes the fun for me. If I have to worry about money while being creative, my creativity is instantly destroyed. This might sadly be the end of Cursor for me if not something is changing soon. Not everyone can afford a 200€ monthly payment, that is basically more money than the cost for my leasing car, just insane.
Sad, but I am already looking for alternatives I have not much options. Will try Claude code and am actively looking who is giving me a better offer.
i been waiting almost a month and one fixed my account i have a student account and it been charging and payments in someone else name and no responds and not fixed yet
ok nice thanks, yeah i have been burning through requests quite fast it feels on copilot, it feels the agents are not as smart as on cursor, making lots of single line edits where cursor does it in one go.
Either way i feel more comfortable with copilot now, the 40$ should be ok for me, and at least the “free” tier you know exactly what model you get. It’s also a lot more obvious what your usage is and how long it’ll last.
Hopefully copilot agentic keeps improving as it’s very new for now.
I’ve cancelled my cursor for now, maybe i’ll return in a few months and see what the situation is, as it does feel like the best product right now, but i just can’t work with a product i have no idea on how much work i can get done without incurring huge charges.
Wait the $60 package is also now useless? LOL. Actually this is lame, because the frontier model Claude Sonnet 4 and his big brother are expense, because they can actually code properly and think. You can’t go back to older models, so is the only package worth it now the 200 bobbins one??!
Let’s be honest — your explanation doesn’t hold water. It’s internally contradictory and honestly feels like you’re just making excuses.
You claim that “most users’ costs have stayed fairly constant” and that “the vast majority of Pro users don’t run out of the included $20 usage.” Okay — so if that’s true, why the hell did you feel the need to change the pricing model in the first place? If most people aren’t even hitting the limits, then clearly the system wasn’t broken.
Then you pivot and say that some requests now cost significantly more — fine. But you already had a system in place for that, like Sonnet counting as two requests. If a small number of heavy users were pushing limits, you could’ve adjusted the request weights or added premium tiers. Instead, you decided to punish everyone by switching to a complicated, opaque token-based system that nobody asked for.
And the “unlimited usage” spin? That’s just misleading. You made it sound like Pro users were getting unlimited access — but it turns out that only applies if you use Auto, which randomly routes to different models and gives users zero transparency. No control over which model is used, no clear idea of how tokens are spent, and yet you still enforce a $20 cap behind the scenes. That’s not “unlimited,” that’s a bait-and-switch.
What’s even more frustrating is that you had a balanced system. Light users who didn’t use their full quota helped offset the cost of heavier users. It worked. But instead of keeping that healthy balance — without increasing your actual costs — you chose to flip the whole model and offload complexity and risk onto your users.
Don’t try to sell this as “fairer” or “more accurate” pricing. It’s not. It’s just a worse deal, wrapped in nice-sounding language, and it insults your users’ intelligence.
In itself, it is not useless, idea is OK. What makes it useless (what makes also 20$ useless) is 2.5m tokens spent in a single request due to Cursor/model/whatever cowboying something you didn’t ask it to do and not doing something you asked it to do. Combination of things that can use your tokens is currently just insane. And you get what you asked exactly for once in a five tries.
Have you tried it? Seems very very good actually! I’m with Windsurf now and I’m enjoying it a lot, more than Cursor already. But this Kiro seems very good also
Yo! I`ve installed Claude Code (CC; with pro plan) in my cursor. And working at same task - refactoring JS into my project with Sonnet 4 (In cursor it was -thinking model). This is 2 screens of usage CC and cursor. Yes, I know that I did not change the model to thinking in CC and this is not a very representative comparison. But still! Should so many tokens really be spent on thinking and cache?
Cursor unfortunately will not survive in this AI IDE run. AWS just launched their IDE Kiro. Kiro Pro: $19 - 1000 requests, Pro+: $39 - 3000 requests. For now, only Sonnet 3.7 and 4 are supported. So it even more interesting than Copilot!
