why do i get charged on auto mode?
I am very confused and worried now, i feel i can’t trust the decision to depend on cursor alone! Cursor is great, but the pricing just confuses the hell out of me; i sure i will not be able to afford this long term!
2 Likes
me too is so much expensive. Please look again the pricings maybe you can do better
1 Like
Fully agree. I wouldnt mind the usage-based pricing since I am a very heavy user and it does produce results for me, but nowadays using claude-4 sonnet with cursor feels more like a bug. I ask a simple question or point the model to an exact error log, and 30+ read and greps are performed, followed by hundreds of lines of modifications I had never even asked for.
I suspect this plays the biggest part in the token draining, a total over-consumption completely disregarding the user instruction that was actually placed.
I love cursor for the UX it provides me to develop, I tried windsurf and quickly came back despite having to pay 85 dollars this month already, but I need something to change, otherwise I feel like I am being taken advantage of paying for AI doing work I didnt ask for upon my codebase.
i agree here, there is something about the UX on cursor that I prefer much more than windsurf, however the pricing on claude 4 is a joke now.
this was a fresh chat, with no rules, and I gave it only 4 files to read, with the total edits being less than 100 lines.
keep in mind the files were not large at all, and sadly auto can’t seem to follow simple instructions and fix simple problems, so I get stuck in the loop to pay for claude 4 usage based or not be able to get through these tasks.
1 Like
The best solution i think who we have is just to cancel our plans. And lets whait for the reflection from the Cursor Team. I think we can find to many others ways for AI development tools.
Who agree my argument please react.
4 Likes
I’m also very confused. last month I racked up supposedly $186 of api usage included in my pro subscription before limits kicked in, when they did I could switch models or wait a day and it would work again and keep adding to that $186, this month, hard stop in 1 week of on and off usage with only $86 of api usage and no replenishing or allowance when switching models. Just a hard stop.
I also noticed the option to opt out of the new pricing model was removed from my account settings when it previously appeared under “Advanced settings”
I’ve got 5 friends using and paying for cursor on my recommendation and have already received calls asking me why I claimed it was unlimited usage at a slower pace past a certain point. Feel confused and upset. Just want some clarity. I racked up $30 of usage in an hour when turning on the usage based pricing which is not affordable for me so i’m left with auto mode which seems to mess everything up in my codebase and not follow rules.. worse yet they won’t even show you which model auto is currently using…
2 Likes
Switched to copilot. If they do any changes in pricing then maybe I will go back to cursor.
Let me reply to a few of you again:
@muratciman If you hit your limits and upgraded to Pro+, and then again hit your limits before the predicted date, it likely means that your usage may still be higher than Pro+ allows for. I’d recommend looked at my previous post with some tips to help get more usage out of your plan!
@banth31 My last post is a start on this, but the team are focused on adding some better in-app visibility on token usage to help see exactly where your tokens are being spent with each request! Regarding your comment about your usage, we are already thinking about this internally, and will hopefully improve this for you soon.
@dhd Especially on this forum, it’s easy to get the impression that the new pricing is a failure and doesn’t serve the users’ needs. However, only a small subset of Cursor users engage on this forum, so it does not accurately represent how the majority of users use Cursor. Our pricing plan changes were made based on the numbers and statistics, balancing the line between ensuring the plans are sustainable for the future while allowing people more flexibility in how they use their usage.
Regarding the 20% uptick, Cursor does a lot of hidden work behind the scenes to power each request, more than just the single request to the LLM you choose with the request you type in. This means we do have costs to pay for. While we do currently have a margin on API usage outside of your plan, we are reevaluating this at the moment.
@Xernive You won’t see the Auto option if you have any API keys enabled. Check your settings as you may have one enabled.
@jiahaoChen @PabTorre You can email us at [email protected] for any support here!
@Obed0101 @0xHACKS While I appreciate your desire to raise your concerns with the new pricing plan, we want to allow people to do this in a way that doesn’t cause other posts (bug reports, discussions, feature requests) from getting buried. While we want to make sure voices are heard, we will hide and delete posts that aren’t constructive to the conversation.
@holden Sorry to hear that you’ve not had a great experience so far with our plans. Looking at your account, you’ve consumed >$1000 of usage so far in this billing month, which is a significant volume of requests! The usage projection is a rudimentary estimate based on your trends so far, but as the system is still new, it only has so much data to go off so far! I’d check my earlier posts for some advise on getting more usage each month, but my biggest recommendation would be to not use thinking models as often, as you might be surprised at how well the non-thinking models perform!
@Aleks_Sergeenko Just to clarify, you do still get unlimited use of ‘Auto’. We have since removed the ‘Unlimited’ wording from our pricing page to avoid any confusion. Besdides Auto, your usage is dependant on the size of threads with the AI, including messages, files and rules.
@Theus-Gui-Developer I apologise if my message comes across as throwing something in his face. The goal by sharing this is to bring visibility to it, as many users do not know the costs of LLMs via the APIs directly and can often be surprised at the value they are receiving. No private usage data was shared outside of a rough $ figure that is indicative of their usage. While our marketing and rollout of this pricing change was sub-optimal (see here), we believe the new plan is more flexible and beneficial to the majority of users. The users on this forum are a small subset of our total userbase!
@mattj We aren’t currently aware of a specific bug but the team are investigating if there is an issue with Cache Read token usage specifically. If an issue is found, we will ensure to make things right with those affected!
@em00k Hey, to clarify, we still offer unlimited usage of Cursor within our ‘Auto’ mode, which was what we initially referred to on our pricing page. We have since changed the wording of our pricing page to avoid any confusion, but the plan hasn’t changed since it’s announcement. I do apologise that we have lost your trust, and we hope to gain in back in the future 
@bsmi021 Hi, to be clear, our pricing has changed once so far, although our wording and communication has changed to be clearer. To summarise, users get unlimited ‘Auto’ usage, and (for a Pro plan as an example) at least $20 of LLM usage each month (this is a good way higher in real usage). As I’ve described in posts elsewhere, Claude 4 is a very token heavy and consequently costly model, and trying to price that under the unit of “requests” was difficult enough, but pointed to a more difficult future. On the old structure, we’d run the risk of having models at 5x requests, 10x request or more. Additionally, the concept of a “request” was unbalanced, charging a user the same for a one word query as it would for a huge query with tons of files. The new system is a big change, but one that means a user’s usage is accounted for, and they are not punished for using requests on smaller models if they choose to be more conservative with their usage. On the other end, it also means that, as models get increasingly costly, we can continue to offer them within all plans, as the consumption of a users monthly usage matches the cost of the model to query.
@Sheikh_Zaid I’m sorry to hear you are unhappy. Any requests that error should not take a hit on your usage - you can check this in Cursor - The AI Code Editor. If you continue to see errors in your requests, please post in Bug Reports and we’ll be happy to help you there!
@bsmi021 We understand that, for some users who made heavy use of Claude 4 Sonnet, the new structure does provide you with less requests than the old plan - however, the cost of Claude 4 requests were more than could be counted under “2x requests”, and was a hint towards a very confusing future of having models that consumed 2, 3, 5, 10+ requests. Also, the old structure meant that some users were paying for others - those who said ‘hi’ were, in essence, paying for those who submitted a bunch of files and complex instructions.
@wtester While those on annual plans were automatically moved to the new pricing, anyone who opted out via the web dashboard will remain opted out until the end of their billing year. Anyone who missed the boat can email us at [email protected] and we’ll help you there!
@yunuss5461 You do still get unlimited usage of models via the ‘Auto’ selector, which has no usage cost. However, we found this to be a confusing wording, so changed it to ‘extended’. This is still true with our Pro plan, as you will always get more than the price of the plan in model usage each month, compared to the API costs you would pay for doing the same requests yourself via API keys.
@Illuminationx We are currently looking into this to see if there is an issue with the count of cache read tokens. If there is, we will ensure to make things right, but we are not yet aware of an issue here.
@wtester I was still writing this post after my last one to answer individual questions.
@anonix98 You aren’t getting charged, we are just showing you that you have used $0.35 worth of API usage.
lol yes, at one point i was asking the auto mode what model it was, and it would keep saying cursor has blocked it from saying what it is… seems like they aren’t probably using the auto models they claim to.
Now it always just says ‘based on claude 3.5’ , but the way it is dealing with code is definitely not how claude 3.5 is, it couldn’t even fix / do simple tasks without going off track, even in fresh chats.
if you ask it now over and over, it will keep saying it doesn’t know what model it is, then claude 4, then 3.5, then back to it not knowing what model it is again… seems like it definitely isn’t even a decent model.
it would seem the auto model is being hidden on purpose and is not the model they portray it as, because it is definitely not claude 4
4 Likes
@danperks perhaps you can explain to me where in the below image of the email that I received when I subscribed to Cursor, am being told that I’m limited to $20/month. I’d love to know why Michael Truell didn’t include that in the email to me when I signed up for Cursor. I must be stupid for expecting the level of service that I subscribed to.
Your ability to provide additional models in the furutre is not my problem, I paid for and expect 500 Premium Requests per month. Cursor making or losing money off of me is not my problem.
Please advise if I can get a prorated refund for the remainder of my subscription period as clearly Cursor will do what Cursor needs to, and ■■■■ customer service, amirite?
@danperks it seems even if there is “no cache read bug” that the cache read tokens are so high that even if the product is working as intended that there needs to be change made. I am not very bright, but I do not see how the huge uptick in cache read compared to CC is needed. now obviously in CC I cannot view cache read tokens but comparing cache read to cache read is not really the point, it is dollar vs dollar.
1 Like
@wiz1 The model is not told any specific instructions in Auto, so it’s entirely down to the model that Auto chooses to decide how it wants to reply. We neither instruct it to censor itself, or specifically tell it its own name!
@bsmi021 I do apologise for that email. The email is now changed to be more representative of the current plan. If you are signed up for the yearly plan, feel free to reach out to [email protected] to be moved back to the legacy plan for the remainder of your billing period!
@banth31 I agree that Cursor, especially on the new pricing, needs to now be more efficvient with it’s token usage, especially if there isn’t a bug here. The team are focused on this and should have more to show for this in v1.3 (coming soon) and v1.4 (not quite as soon!)
That is definitely false as I asked it a while before today (maybe 2-3 weeks ago and it clearly stated that cursor has specifically blocked it from saying what it is.. and agreed it was dishonest. If I knew about the forum today I would have had it screen shotted to send here.
And before the change above, it would always give the correct model, sometimes it was gpt4, sometimes it was claude 3.5 … and it would actually work correctly and solve the problems properly, unlike now where it struggles to do anything correctly, even when it really is simple tasks that 3.5 / 3.7 completes pretty easily.
Can you tell me what the model list is for auto? I think it is fair we at least know that, we should know what we are working with.
Its not just reads its writes too. I am getting 72,000 cache reads tokens being used up from saying test.
5062b4d3-f9e5-482c-b452-5ca228fe2bc7
Thank you for this feedback, however even when I take all this into consideration, a simple request, all together including rules, memories, context, mcp etc is still insane - For those freaking out - it has to be a bug! - #26 by liquefy
edit: just had a thought, is some form of a folder structure or anything attached to chats automatically? even if i dont include it? or some data that has been indexed? keep in mind this was a new chat and no context attached
I also did that change and even though Claude Code is far superior at helping with the code, it does not have the “Restore Point” feature that Cursor applies with each message (as far as i know). I am genuinely curious, how did you cope with that?
Thank you, @danperks, I have written to that email address and will wait for them to get back to me.
Took them 4 days, good luck
esc twice