You lost me as a user. I used to use Cursor in my free time for my hobby, in evenings or weekend or holidays.I used to pay for it and never reached the limit of 500 fast requests, which means I paid more than $0.04 per request. But I knew I had these requests when I need it.
Now you have request limits, the real cost for the models plus interest, and still ask me to pay $20 which will be charged as the model price. So I started getting less than I could before for the same price, while the price model has been switched to API one, as if it’s claude or openai. I don’t know the person who implemented that, but whoever it was they should have thought before offering these things to people with a math background.
13 Likes
- We were told that the opt out button being removed was a bug?
- Erodes the trust 100% more than it already has been.
- We have no idea anywhere what each model/token costs.
- Previously we could see how many requests each model would cost.
- Why can’t we see anything (even a bar) showing us:
- How much usage we have left
- When usage will ‘restock’
- Making it VERY clear when usage-based pricing kicks in
- How are you still telling me I am getting better value when I am getting less than half the requests (its closer to 1/3) than I was previously?
The new way of pricing is so unpredictable. It is so hard to determine if Cursor will even be usable.
3 Likes
I’m gonna bail for now. I never really had any issues, never hit any limits or anything. I don’t think I use AI in the same way others do on this forum. But the lack of transparency is a deal breaker. I need to be able to see where I’m at with my token usage. I’ll keep checking back to see if you guys ever figure this out. I actually like Cursor. Good luck.
6 Likes
I reverted back to the old pricing method , but now suddently what they dont tell you and which is not obviously clear from the UI perspective is that they charge you 2x request for sonnet 4 , its not stated anywhere in the UI ??? , unlike in Windusrf where you select the model it will clearly state it consumes 1.5 or .25 of “credits” or “requests”
2 Likes
Yeah, at this point I’m just waiting for Cursor to offer me a ‘free spin’ instead of coins, it’s just a few questions and suddenly my Pro plan is over 
1 Like
That’s exactly the challenge I have, claiming $205 in API cost for what is seemingly 400k input and 132k output tokens doesn’t put it anywhere near that ballpark. Even with the 20% margin on top. So unless there is something in the cache tokens (which they have always said its just input/output) I can’t see how this actually works.
You guys have best Tab model, but I am not paying 20$ just for that.
Keep the 500 request mode, otherwise you would loose more than half your customers.
2 Likes
Hi,
I am extremely disappointed with how this $20 charge was applied to my account without clear understanding on my part of what was activated. I did click on an option that I did not fully understand, and it immediately charged me without any clear warning or an additional confirmation step before taking the payment.
I never knowingly authorized this type of automatic token-based charge, and I feel misled by how this was implemented without clear alerts about limits or additional costs before the charge was processed. This is a breach of trust and creates insecurity about continuing to use the platform.
Additionally, I noticed that the 500 premium model requests per month (Claude 3.7 Sonnet, Gemini 2.5 Pro, GPT-4.1) that I was supposed to have are missing, while the plan stated I would have unlimited requests on “slower” premium models. This is extremely disappointing and does not align with what was advertised.
I am also a Cursor subscriber and noticed that after around 30 uses, I was prompted to purchase additional tokens. This was not made clear at the time of subscription, and I am frustrated to have paid for something that quickly became limited.
Previously, I used Grok for six months and made countless requests without ever facing hidden limits or surprise charges. I am seriously considering abandoning Cursor and returning to Grok due to the frustration and distrust this situation has caused.
Given all of this, I am once again requesting an immediate refund of the $20 charge and for usage-based billing to be permanently disabled on my account to prevent further incidents. I also request a clear and detailed explanation of exactly how this happened so that I can avoid these issues in the future.
I look forward to an urgent resolution.
Best regards,
Mateus
3 Likes
I just hit my limit AGAIN and it said I saved $197 but, in reality, I’ve spent probably 3 - 4 times as much as I normally would have on the old plan and I’m only 1/3 of the way through my monthly plan. The math isn’t mathing…
Sorry Cursor. I’m paid well but not well enough to spend this much on a solution I can get (almost) for free elsewhere. You product was amazing. Someone somewhere made some pretty terrible decisions and things haven’t worked out for anyone. I’m sorry to say it, but I’m out.
As a side note, I just wanted to say, I feel for the developers, designers, and product people who had to work on this initiative. I feel for the forum moderators and support staff who have to deal with the angry customers. I feel for the finance team who have to deal with all the refund request. I hope you’re ok. Fingers crossed, this will be resolved in the not too distant future and hopefully Cursor will rectify the situation and things will return to normal.
5 Likes
Reach out to support and ask for a refund.
1 Like
I personally find it increasingly hard to feel for the moderators, ambassadors, and support staff, across Cursor communication channels.
They’re actively taking part in creating an echo chamber in which “everything Cursor does is right”. In my opinion it’s partly responsible for the unfortunate decisions Cursor higher ups are making, and their inability to properly reflect on those decisions, hiding behind “we didn’t communicate well”.
I wish they worked less as “advocates for Cursor” and more as “advocates for the community and consumer rights of users”.
4 Likes
They’re in a pre-bankruptcy state, and either one of the major providers (xAI, Anthropic, Google) will buy them out soon, or in a short period of time, they’ll be worth nothing.
Over the next 6 months, a significant leap in AI is expected, and every serious provider will be expected to have their own IDE with a dedicated user base and special offerings.
Small intermediary players who don’t provide their own models simply won’t survive — they’ll be crushed by the big players.
5 Likes
I would love to just top-up $20 for 500 more requests than pay as you go. I think that was an option before, can we have it back?
4 Likes
Hey all, lets answer some queries so far:
Rate limits is a confusing way to describe the system, and our latest blog post about the pricing does a better job of explaining this, but I’ll summarise - each month, you get a level of credit that is >= the cost of your plan. As you send messages in Cursor, you will consume the credit at the same rate you would via the LLMs direct APIs.
Think of it like pre-purchasing meal credit at Disneyland - you pay upfront for your plan, and you can spend that credit at any of the restaurants available in Disneyland.
On top of this, you also continue to get unlimited usage of Tab, and now also unlimited usage of the ‘Auto’ mode, as we can intelligently route users to frontier models based on capacity to help give you more for your money.
While the option to opt out of the new pricing was available via the dashboard, all users were moved over automatically as for many, they would see no difference their usage and would actually get more requests from their subscription each month, depending on their usage.
You can also still send us an email to [email protected] and we’ll be happy to move you back to the old plan either until the plan is fully sunset in the future, or until your the end of your billing period, whichever comes last.
We appreciate the sentiment here, and that our efforts to resolve this have regained some of your trust here! Regarding our subreddit (and this forum), we do currently use a mixture of Cursor staff, Cursor ambassadors, volunteer moderators and AI-powered moderation to try to keep the forum and subreddit as constructive places for people to interact about Cursor.
During this period following our pricing changes, we have been trying to maintain a good balance of constructive criticism while trying to hide or delete posts that are inflammatory and not beneficial to users or us. While this may seem like a kind of censorship, we try to ground these decisions within our community rules.
I’ll make sure to feed this back to the team to see where we can improve here moving forward.
I apologise that our experience with Auto has not been great so far! Auto will always route between a frontier model, like Claude 3.7, Gemini 2.5 Pro and GPT 4.1, but some of those models have better or different behaviours to others.
This is exactly what our usage-based pricing system is for! If you do not want to upgrade to a higher plan, you can set a spend limit and allow Cursor to bill you directly for each request made outside of your plans allowance. If enabled, this would takeover seamlessly (with in-app alerting to tell you its happening), and you can always check and configure your usage-based pricing on cursor.com/dashboard
We agree, and this is where we missed the mark. We were not clear enough that ‘Auto’ was unlimited, not all models and usage inside Cursor. We hope this new structure will actually give more requests and usage to our more price-sensitive users. Regardless, we were not clean enough with users on this.
While the speed of responses can fluctuate based on the demand (Anthropic responses slower when it’s busy!), there shouldn’t be a major change here - this seems like it’s more likely a bug, as excluding the ‘slow pool’, we’ve never had any kind of artificial delays or queuing on requests.
If you continue to see this on the new plan, please post in Bug Reports and @ me so I can help you directly there!
While the pricing change, buy-and-large, is an improvement for the majority of users regarding how much they get for their money, those who relied on the state of the art models consistently will likely have seen a cost increase. Offering Claude 4 Sonnet at $0.08/request was not a sustainable pricing model and the new system should now be much better at tracking your actual usage fairly - small requests, and small models will allow you a lot more than 500 requests, bigger extensive models will use a lot more.
I would recommend opting for our Pro+ or Ultra plan, as the additional usage you get with those plans is substantially higher than you would get on Pro with usage-based pricing enabled.
I agree that transparency is best there. We have already added in-app alerts to tell you your progress through your usage, and also included the table on cursor.com to show you your exact usage, but we hope to improve on this more as time goes on.
Looking at these numbers, these do seem to be correct. The exact figures are hard to work out, but the maths would put your cost somewhere between $172.50 (using the 5 min cache-write rate) or $236.70 (using the 1 h rate) for the cost of Claude 4 Opus Thinking.
I’m sorry to hear you got caught out here! Please drop us an email at [email protected] and the team will be happy to refund you for any charges you may have received due to the automatic migration.
Moving forward, it seems like you’d benefit a lot from either Pro+ or Ultra, as you will get much higher usage with those than you would on Pro and using usage-based pricing to reach the same final $ amount.
Please drop us an email at [email protected] and the team will help you out there!
The old plan still has the slow requests pool as before. The only difference to the old plan for those who have opted out is usage-based pricing, as additional requests are no longer charged at a fixed $0.04/request cost.
If you have emailed [email protected], we are seeing a ton of inquires right now, but the team will get back to you as soon as we can!
You can read the full details about the system on our blog post here (Clarifying Our Pricing | Cursor - The AI Code Editor) or on our docs here (Cursor – Models & Pricing).
While the option has been removed from cursor.com/dashboard, please drop us an email at [email protected] and we’ll be happy to help you out.
You will see in-app alerts when you are approaching your limit - we’re hoping to improve the visibility here further too!
3 Likes
Hey really appreciate you taking the time to go through and provide so many responses.
For mine about the pricing of OPUS it would be really helpful to have a simple api price list, showing input / output / cache cost of tokens. The big lack here then if the pricing stacks up roughly from your perspective is that as the user we have no visibility to it really.
Can we like… stop pretending that Claude 3.7 or GPT 4.1 are “frontier models”? Or that we get something else than GPT 4.1 more often than once a blue moon?
Why not just call GPT 4.1 a base free model like Github Copilot does, and drop this illusion that auto mode is some kind of premium feature.
Please.
6 Likes
1、([email protected]) Reply: The option to opt out of the new pricing model is no longer available.
2、You sneakily swapped it for some ridiculous new pricing where every reply burns max tokens — and after a few tries, I’m locked out for days!
3、And you still have the nerve to say you’re saving me hundreds of dollars?!
4、And don’t even mention “Auto” — it’s barely usable, just a toy!
5、Back when you were tricking users into purchasing, your documentation clearly stated: 500 fast requests and unlimited slow requests!
4 Likes
I don’t really understand what I’m getting. I was notified that I had used my monthly credits, and then I think I got charged (an extra, random amount?) for some usage, but I assumed that unless I did something special (didn’t there used to be glowing brains that indicated pro models?) I would just have slower/stupider models to use. It should be obvious in the UI if you’re using PAYG credits above the $20 p.m. ones. Thanks.
Hi everyone,
I’ve been using Cursor for a while now and was so impressed in the beginning that I purchased the Pro subscription for a full year just a few days after trying it. At the time, it felt like a no-brainer.
However, with the recent changes to models and pricing, I’m finding myself more and more disappointed. Today, I uploaded a screenshot into Claude.ai (free version!) and with just two prompts, I got a fully working 1:1 (about 80%) landing page. When I try something similar in Cursor Auto now, it just feels broken. It’s heartbreaking, honestly.
As a Pro user — and I believe many of us here are experienced developers — it’s frustrating to feel like we’re being left behind. The model quality has dropped to a level that makes it borderline unusable for many tasks. And to make things worse, sometimes the models are simply paused altogether.
I was seriously considering upgrading to Ultra next month, but I’m now rethinking everything. Claude with code + terminal access is looking more and more attractive — even with its current limitations.
Cursor is such an incredible product with so much potential. I just hope the team reconsiders the current direction. Many of us really want to stick with it — we just need the performance and consistency back.
7 Likes