Look at what that question mark says when you hover over it. It says not all models or not all modes. Auto is unlimited. And auto does use sonnet 4 sometimes.
@tumeden We decided to change the wording here as unlimited wasn’t a clear way of describing the new plan structure. You do still get unlimited requests by enabling ‘Auto’, which is free and unlimited to use, and routes your request to one of the frontier models (Claude 3.7, Gemini 2.5 Pro, GPT4.1).
Your usage is now based on your actual LLM consumption, so if you are seeing requests that use a large amount of tokens, I’d consider starting a new chat thread to avoid too much of your previous conversation being unnecessarily sent to the model!
@dhd Individual requests should not by themselves take up a large amount of tokens, but as you continue to go back and forth with the Agent, that is when the conversation gets increasingly long. Each message you send also sends the previous conversation history to the model, meaning while your first message may use 5k tokens, your next message will then be 10k tokens, as the 5k from before gets included - this stacks up quickly if you stay within the same conversation for too long!
@carlosmartinpavon We now have in-app alerting for when you hit certain milestones in your usage, but we are still working on bringing better visibility to this soon!
@Gigantoad I can see from your account you’ve used over $150 worth of API requests so far (you can see this on cursor.com/dashboard), so I am not surprised to hear you are approaching your limit. It seems like your requests to Claude 4 are larger than average, so I’d recommend trying to restart your chats more often, to ensure irrelevant messages and files arent repeatedly send to the model!
@emitter You can see all the up to date info on our plans here: https://docs.cursor.com/account/pricing
@Codingbotguy I’m surprised to hear you have managed to consume enough to hit our limits, although looking at your account, you have used >$2000 worth of model usage already! Thats over 15,000 requests to models (more than 1 every 2 minutes, 24/7). I’d love to hear more about your usage pattern that has got you to this point!
@dhd I’m sorry to hear you have cancelled with us. For reference, you can see your exact token usage for each request on cursor.com/dashboard and, if you want, confirm the pricing with the API costs of the models themselves.
@weiming_hu Sorry to hear you are looking to cancel your subscription. If you email us at [email protected], we’ll be happy to help you out here!
@banth31 If you visit cursor.com/dashboard, and switch to the token view, you can see exactly how many tokens your request used and cross-reference that with the API pricing for the model you used to see how that number is calculated!
@0xHACKS Sorry about that - I have relisted your thread!
@shijunti19 While I cannot see the exact details here, I can see you had some very large conversations with Claude 4, which may have caused you to consume your usage much quicker than average. I’d recommend restarting your conversations more often to ensure off-topic messages and files are not sent to the model when not necessary.
@Akeem44 The new pricing plan is based on your token usage - longer conversations and more files sent to the model will consume your monthly usage more!
@dreamrandomhub We are hoping to improve our visibility soon on what is making up the tokens that go into the model, but besides our system prompt that we attach, all other context is generated by you - to summarise, the input tokens will be made up of your conversation history, rules, memories, attached files (included older ones in the thread) and your own prompt to the AI. If you are finding your token usage is high, trying to keep those areas concise and on-topic would likely go a long way to help!
@Jaeder I appreciate the sentiment here, and want to assure you this is a priority internally. As tokens is not the unit of cost for a request, we are working to bring more transparency to what is accounting for the tokens sent to the model!
To everyone seeing excessively high token usage, we do want to make sure everything looks correct here.
If you can, sending a Request ID with privacy mode disabled allows us to see what has caused the token usage you are seeing, so if anyone can send one over, we’d be happy to look into this to confirm everything looks correct, and provide some specific tips on how to reduce the token usage in future messages.
We are also working to improve the transparency here - while you can already see your token usage on cursor.com/dashboard, we are hoping to bring more in-app UI to help you see this before you send the request, so you can try to make your requests more efficient!
4 Likes
I can assure you that is not the usage patter i hardly code for 5-6 hours a day. Also almost every one has over 2-3k model usage in their screenshots so not sure why the surpirse.
1 Like
I completely agree with you—this situation is frustrating.
In my case, I upgraded from Pro to Pro+ not long ago, expecting extended usage as promised. When I checked the usage panel in Cursor’s chat window, it clearly said my Pro+ subscription would last until July 21, 2025.
But today, after barely using Cursor for a few days, I’m suddenly being prompted to upgrade to Ultra just to continue using basic features.
The usage data I have (including lines of agent edits and suggestions) shows that I didn’t even come close to heavy usage. In fact, my activity has been minimal compared to last month when I was still on the Pro plan and used it much more intensively—yet it lasted longer back then.
This abrupt change without transparent communication feels unfair, especially since it seems like new billing rules were applied to an existing subscription that I already paid for. If usage quotas are going to change this dramatically, there should at least be clear notification before charging for renewals or upgrades.
I really like Cursor as a tool, but trust is important—and right now it feels like that trust is being compromised.
If anyone else is experiencing the same, it might help to raise this with their support team. I’ve already sent them an email explaining my case and asking for clarification. Hopefully, they’ll address this properly.
Let’s see if Cursor responds, otherwise I’m also starting to look into alternatives.
3 Likes
Hey @danperks - might be a great idea to get a how-to out there on optimizing your spend. seems like the cursor team knows more tricks than your average user does lol
also I think the solution is to not have 20/60/200 dollar a month tier plans if unfettered usage over the span of the month would be thousands of dollars. seems like a higher dollar monthly plan is needed. Cursor provides a ton of utility, but it is helpful to know how to budget for it. I spent 200 dollars on my month plan which got me 28 days, the other 2 will cost me more than 200 at my current usage.
This new pricing model just doesn’t make sense for most. Costs are up sky high and the response seems to be “well your prompts ■■■■”
Also no response to the 20% uptick for sonnet issue. This will be my last month here.
1 Like
Cursor is great tool and it would be much better in future but this one change is forcing us to look into other options as Kiro and Trae and i believe current cursor token calculation system is wrong it’s hitting limit within 10 days without much uses.
2 Likes
The magic trick is to ask good questions and use auto.
Thanks for the reply, but it’s not really helping. Tell me how these numbers relate to going from projected to hit limit on 26th to 18th and then within 1-3 requests to limit hit when it all happened on the 15th and I don’t see any major case of suddenly having used way more tokens than usual.
Tell me why we can’t get a simple percentage number? You’re calculating this somewhere, clearly. Show me that I have used 70% of my requests so I can get an idea of how much I’m using when the next request makes it jump to 75%. Why not show that instead of these weird projections that feel like something must be tremendously wrong with the math?
This is funny, because I used sonnet 4 only, then got blocked like 2 weeks ago, used like 10 gemini/gpt requests, and now apparently I’m blocked from using these models as well, so much for 550 gemini and 650 gpt requests…
1 Like
Use auto on back ground agents, it is free and unlimited, and back ground agents only use frontier models.
How? I only get the option for thinking models and AUTO is nowhere seen anywhere.
Most of users complain about pricing and tokens. Before I have used Cursor for almost everyday now I am looking for some better option. Cursor become worse.
1 Like
I feel the same way. I also don’t know where to apply for an annual subscription refund. If anyone knows the process, please let us know.
1 Like
@danperks, I’m looking to cancel out the remainder of my yearly plan. Is it possible to get a refund for the unused months?
Could not have put it any better.I won’t be renewing my subscription. Cursor has become almost unusable with this new plan. I am scared to ask it to do anything in Auto mode because it ends up breaking a lot of things.
I can barely use any of the models. It’s like I have hit a limit on every single model. Google 2.5 pro, sonnet 3.5,3.7,4.0 even o3. I literally cannot use anything other than auto.
The Cursor team seems to forget that majority of us are still engineers. If their product isn’t solving a problem, we’ll look for another tool or just write the code ourselves.
Claude code here we come🤘
4 Likes
I’m feeling ripped off too. I prepaid an annual Pro plan, and using Claude-4-Sonnet only lasted about 3 hrs of coding time this month before I hit usage limits. Not happy and I am currently looking for other options. Many forums are filled with people disappointed with Cursor’s apparent bait and switch tactics.
1 Like
Right now I can’t see how many Pro and Free requests I’ve made each month, or my total request usage.