With all the changes on pricing it’s getting tiring visiting this forum from seeing “everyone” complaining. I say “everyone” in quotes because I don’t think the number of complaints is representative of what people really think. But those who are complaining are very vocal, I guess.
Is Cursor a company that does not know how to handle communication? Yes.
Has Cursor made decisions and announcements without thinking it through and planning carefully? Absolutely.
They have to pay the LLMs provider, they may get a discount but they have to run company, so they need to have a gross profit from those API calls to maintain the business.
I think there are two reasons why people are complaining:
Some people may not understand how LLMs work or how agent / ask mode works, token usage, context, real API cost, etc.
Some might be using Cursor as a vibe coding tool more than an AI-assisted IDE.
Do you really want to compare about cursor pricing? Then try something like Roo Code or Cline, put your own API key and do a couple of things in agent mode.
There are a lot of platforms for vibe coding like Replit, Lovable, etc. They are all priced very similarly, and they all have a very similar number of requests per month (~200 per month for 25$). If you compare what you can do on those tools against what you can do with a 20 monthly subscription on Cursor, it’s laughable. Cursor is a much better value for money compared with those products by a huge margin.
I think the way Cursor is charging, exactly as API cost, is really the only way of doing it. Considering the popularity of Cursor and that a lot of people seem to be using it more as a vibe coding tool than as an AI assisted IDE makes the business unprofitable if they charge per request.
Do I wish that Cursor were charging by request instead of passing the API cost? Of course, but I know that that is not profitable for them because 500 requests are very different for someone who is using Cursor as an IDE, giving specific instructions about a function, adding context carefully, etc. to another that are just asking vague questions in agent mode and let it be… (good luck with that).
I agree with you. I need to add that the visibility into usage is non existing on the App. I need a counter or more counters that show me what is going on right in the app. I need to see if my chat session is causing unnecessary usage and I better start a new chat. I need to see how many requests I have in my disposal and how many I have used. I need to see the very advertised saving right away.
I need to steer my usage and plan everything before I get stuck and come here and “rant“ and “troll“ and “complain“.
I am happy with Cursor, and have not found a replacement for it yet, so I am a passenger on this ship, which is changing it’s fee on the fly, and I feel like the owners or managers are afraid of announcing the real price right away, in case users run away and stocks crash and whatever the reason.
Just come forward and say this is the price, no surprise, take it or leave it.
hi @DavidCodex and @cocode thank you both for your feedback. Cursor Team mentioned they are looking into improvements and better communication in app & otherwise. Some changes were already implemented with usage estimate notifications in chat but also more is planned and on the way.
As a note on ‘rants’ I think also that the community would benefit from forum member interactions as there are many users who look for advice or like to share experiences. There are already plans in motion on Cursors side for forum improvements. I would also appreciate your suggestions.
People are complaining because despite paying for a year upfront last month they got the 500 requests they had paid for by the agreement and this month they will get 50 if they are lucky.
You can go back months and read posts about “hey, can we have better visibility of our usage on the dashboard, it is hard to manage”.
If you see this number of tokens and still think it’s just a matter of communication or pricing—really? Cursor has a problem, but instead of fixing it, they ignore it or delete posts.
Massive token usage direct proportional to context used. E.g. attached files, rules, MCPs, etc.
Per your screenshot 3M input tokens on Sonnet 4 means there was more than 4000 book pages of text added in total. Thats equivalent to 15 novels or 20 academic papers.
Note that the token usage is counted as reported by AI providers and therefore shown in the Dashboard Usage.
For so much context used your screenshot shows reasonable amount of allowance received.
The question I’ve asked 3x times to you today, and have yet receive an answer: His token usages, if it’s 100k chat context, and in a single turn to the Agent, it decides to do 5 edit_tool calls and 3 greps is he billed 100k chat context * (5 + 3) = 800,000k tokens?
Because then it makes sense these insane token usages. If the Agent is absolutely unconcerned about uneconomical use of tokens, then ofcourse it would be sending the entire chat context for every single tool call in makes in a single chat turn - and requiring cache reads on everything.
Stop whining, Cursor’s pricing is fine if you understand it and use it correctly. If you’re complaining, you probably don’t know how it works or aren’t using it like a ‘real dev’.
Saying users are just “vibe coding” or don’t understand tokens is a lazy way to dismiss real problems.
Most of us aren’t asking for free AI, we’re asking for fairness, clarity, and models that don’t feel engineered to burn through usage.
Comparing Cursor to worse platforms doesn’t make the core issues disappear.
Sorry there are a lot of threads and I have not yet gotten around to revisiting all of them. a 100k chat context is 100k tokens total not 100k * (5 + 3)
Agent fetches code that is relevant to the task. The same goes for edit tool call, it only does change that is produced by AI based on the request processed. Not 100k tokens.
Cache reads are cached Chat thread requests & responses incl. code etc. They are 10% the cost of input tokens, therefore very efficient and cost saving.
We’re not vibe coding. It’s a tool to solve a job.
Most of us began LLM coding with gpt-3, pasting stuff into a chat interface. Then aigen, cline. Then using file-attachements with AI Chat standalone apps like Bolt.
Cursor’s 1 year ago was completely different. I remember when it’s Agent mode sucked, and you could only use the ‘Ask’ mode. Then the agent mode became better. Then you learned claude 3.5 was too biased on making tool calls and had to switch to gemini-2.5 to manage the x/500 requests per month.
Most of us have been through an Evolution with tool.
And the rage is warranted because it was a great experience for the prior 300 days I’ve used it, and the last 10 days have just been ■■■■. I’m forced to switch to using API keys on-off to budget my token usage. I have to slow down and ration out my tokens. I’m suddenly testing new tools I could switch to like Warp or Claude Code.
Instead of focusing purely on building product, I’m force again to deal with token rationing. It’s just moronic. And I blew through the $20/m budget instantly.
The same way I used claude 4 weeks ago and cost me $2-10 in usage now costs $100?
Exactly. No one’s saying it wasn’t great, that’s why the decline hits harder. Going from smooth dev flow to token rationing isn’t evolution, it’s regression. Enough said.
The only advantage I see in this is that my competitors are hurt by this too. So figuring out how to squeeze the most out of Cursor now is a competitive edge.
@T1000 moving complaints to feedback category as a start, categories should have strict rules and discussions should be only centered on AI, maybe an automated agent that recognizes categories from content and move them to the correct category, too many in ‘how to’ asking questions, bug reports that are feature requests, […] that would already do a lot to forum quality, keeping forum quality means those that quit remain here adding value and maybe returning sometime in future, we just need some cleaning <3
Thanks @normalnormie absolutely agree with you. For now mostly manual processes but AI could be helpful, have to check what’s possible. There are forum improvements already lined up, so I am curious to see them land.
We need to keep a sober tone, and lift this issue with good manner. I know that at the end somebody is making money of this, but we are also making money with it, so the Cursor owners and devs need to be guided in the right direction to benefit us all.
The moderators here are surely under heavy pressure due to high number of posts, and the least we can do is to be honest and clear and respectful, no matter the bug or number of tokens. We must not let the future generations laugh at us .
I think also that the community would benefit from forum member interactions as there are many users who look for advice or like to share experiences. There are already plans in motion on Cursors side for forum improvements. I would also appreciate your suggestions.
.