It’s straight up busted. And no communication about it feels bad. People are blowing through monthly quotas in a few days.
The count here includes cached read tokens, which get included based on all prior input tokens at every tool call.
Cached input tokens are 10x cheaper than input tokens. If you hover over the token count, it should give you a breakdown by token type.
Even if cached input tokens are 10x cheaper, it is still too much tokens. The cost goes from typically $0.10 to $2 without any reason in the prompt or chat length. Therefore, our plan limit is reached incredibly fast. I was typically under 150k tokens and jumping now to 3M…
Do you have plan to look at this issue? Without solution, Cursor is just unusable for us
If you are finind you have a large amount of cache read, this suggests your input from Cursor to the model contains either a lot of repetition, which suggests you might be using an old / long chat with bloated context! You may have had lots of messages back and forth with the AI, or attached lots of files to get to that point.
In this case, you have the power to manage this yourself! Regularly start new chats, keeping rules concise and to the point and limiting the files you attach to only those necessary will all help to make your requests more efficient, and therefore have less of an effect on your usage!
1 Like
The response to everyone finding usage issues is basically “git gud”. Live and don’t learn I guess.
I just stopped using the cursor - that’s a good solution, isn’t it? ) It seems to me that someone in your group has gone crazy, thinking that I should pervert myself and somehow use the Cursor more correctly? Haven’t you thought that the user should simply understand that he pays “so much” and shouldn’t have a headache about “How can I use the Cursor so that its developer can’t deceive me and write off extra money?” You now remind me of a mobile operator, who once could write off thousands of dollars from a subscriber, because the subscriber did not turn off MMS, using SMS :))
Listen, do you even read what people write on the forum or are you just blabbering on and on like a talking bird? Everyone writes that they don’t use Max mode! I spent $50 in 4 hours using Cloud Sonnet 4 with some hesitation! It used to cost me $1-3! And the fact that I paid for an annual account by signing an agreement and then you simply transferred me to some other tariff plan of yours without my knowledge or consent is pure fraud. No one has apologized or explained yet! Are you sitting at a meeting and deciding “Let’s transfer everyone to a new tariff plan now!” Are you sitting in a village or do you have lawyers and other qualified personnel? This is not a zoo, where you don’t have to ask a giraffe what grass it will eat today! You have completely discredited yourself with this step and probably the law enforcement agencies of the USA or Europe will deal with this scam. And now you sit and continue to talk some nonsense that the user should somehow be more careful when using your product and make sure that your broken algorithm doesn’t steal his money!
1 Like
I kinda got triggered by a message in another post and ended up crunching some numbers. Just leave it here (there’s a table inside).
And I still don’t get why people keep whining and using the third most expensive model when there are cheaper and better options out there 
The cache gets read after every tool use by the Agent. Claude seems to love making single-line edits (at least, I’ve seen it happen a few times). So instead of submitting one chunk of code and reading the cache once, it might call edit_tool 10 times, and you end up with 10 reads of an ever-growing cache.
1 Like
I would spend 5k a month on Cursor if I knew exactly what the 5k a month was getting me. it’s the unknown aspect of the pricing model that is the concern
You pay $200 for the Ultra plan.
They guarantee you’ll get at least $200 worth of tokens. I already got $142 worth of tokens for my $60.
I’m not sure where it’s stated, but apparently Cursor also sells you tokens at a ~20% markup. In return, you get the best agent-oriented chat on the market. Buggy, yeah, but still more functional than the rest, so it’s bearable.
Yeah, it’d be nice to have something like a fuel gauge. But that’s definitely not a reason to stop using it.
I think you’re missing the point. People are upset about how little value the money is getting them.
2 Likes
If I weren’t so dumb and hadn’t wasted my tokens on my own open-source stuff, I’d be making at least $60 a day. And I live in a country where the average salary is three times lower than in the US. Cursor isn’t some charity startup. If you’re making money, you need a serious, high-quality tool. If it’s just a hobby or you’re not earning enough yet, go with something simpler (don’t pick the priciest models on the list).
1 Like
value rocks. I got 1100 of usage at the 200 ultra plan. issue is Claude code is still unlimited with a rate limit, if cursor were marking up that model I would have no issue
Try to change your User Rules with GitHub - Artemonim/AgentCompass: A comprehensive set of rules and guidelines for AI-assisted development in Cursor IDE
Try to work on my How To Optimize Your Usage: The Best AI Models to Use, version 2
Try to prompt- and context-engineer better
Then why the hell are you whining here if you can just use Claude CLI?
Maybe because Claude CLI isn’t that good to fully replace everything Cursor has to offer?
You could even use Google CLI — it gives you 100 free requests a day. I even downloaded it myself. Never used it though, 'cause using a GUI is just way too convenient.
Do you like it? Well, then work in Cursor, why are you trying to convince people of something that doesn’t exist here? 
Or were you asked to write this nonsense here? And I already work in another IDE with transparent pricing, where there is no mugging with a knife at my throat and an unexpected $50 charge for 2 hours from my card!
1 Like
Because I want to see something useful for myself on the forum or share knowledge with others. But right now, every other post is like “I want to use a model I can’t afford. Cursor bad.”
I honestly have no idea how someone manages to burn through $50 in a single day unless they’re running in MAX mode non-stop. Sure, I mostly work on my own projects, but these are the kind of projects that even Gemini and Grok struggle with. And still, no crazy overuse.
Though recently, I did start tweaking someone else’s repo for my own needs—and guess what? Still no wild spending ¯\_(ツ)_/¯
1 Like
I use Kiro from Amazon. It suits me and it has an amazing specification and hooks function. And it is free now, then it will be $40 for 3000 requests. And I think in a month or two it will surpass Cursor. Haven’t you realized yet that Cursor is dead? You can give it credit, it helped a lot, but it will never pay off investors’ money, because giants and providers themselves have entered the IDE market, who do not need intermediaries! Now there will be a war of IDEs and agents from OPEN AI, Microsoft, Anthropic, Meta, Google, xAi and so on