and if it’s not a bug, I just wasted 20 cents.
17 Likes
If it’s truly a bug, which would make sense - then MASSIVE refunds would have to come to save this.
9 Likes
I used to use usage priced all the time. it was never this expensive
2 Likes
good news. cursor needs to be fixed this. otherwise, they will lose most of the users.
2 Likes
Can you give a request id for this message? I can take a look
Also, is this the first message in the conversation, or were other files attached?
If it’s not the first message in the conversation, then it could incur token costs from prior messages in the conversation
I did a few request last night after the limit just to finish what I was working on and 10€ expense, it was literally 10 mins of working, it was NEVER this expensive before, ■■■■
1 Like
Yes, exactly — I checked that too by asking, “What model are you?” and it gave a brief answer, yet the input context in the stats was huge.
Basically, it’s inflated — there are lots of internal rules and hidden calls inside that artificially increase token usage.
2 Likes
I need to keep working but I can’t expect to spend 100€ bills like nothing what should I do
I’ve taken a break and stopped working — I’m just watching the chaos unfold.
Yes, Cursor is the best in terms of user interface; the other solutions are unbearable by comparison.
But Cursor’s policies have changed, and it’s no longer feasible to pay for every little thing.
I’d recommend waiting — I believe they’ll be acquired soon, the leadership will change, and things will get better.
4 Likes
would lovely to be able to stop, but I have tight timelines, like really tight at the moment, I’ll have to see other options out there
I did some math, and yeah, they’re absolutely killing it. (apparently off the bug 
)
Cursor AI has over 1 million users, with around 360,000 paying customers. According to the source : 10 Cursor Statistics (2025): Revenue, Valuation, Competitors, Funding
Based on recent data, they’re likely bringing in $10–17 million per month just from usage-based billing, depending on how much devs go beyond their plan limits.
I could be entirely wrong! 
The only bug is that we keep using cursor. Cursor is dead
3 Likes
Token usage is 100% inflated and that’s with every single model, independently of which mode you’re using. It’s definitely worse in Agent mode obviously.
Pretty sure it’s NOT a bug, and it’s the intended behavior because it’s been like this ever since usage based pricing has been implemented - which coincidentally was the moment they were having their funding round 
. Less and less control is left in the hands of the user with each new update/feature; and with new very cheap SOTA OSS models and other IDEs getting better like ROO, it will just make sense to use your own API key and switch away from Cursor. Using KimiK2 + MinimaxM1 via openrouter to test has cost me less than 20 cents coding a project that took more than 1 million tokens just PLANNING with sonnet 4 (and the planning discussion wasnt even that extensive). I ran through all the included usage and don’t feel like putting more money this month (a mistake I did previous months) so I switches to openrouter + KimiK2. screenshots below.
And don’t dream about refunds lol, been through that already with support a few weeks ago and guess what … nothing “the team will get back to you but please know we’re receiving more support requests than ever so it’ll take time” type thing.
001/
├── calibration/
│ └── init.py
├── core/
│ ├── agents/
│ │ ├── base_agent.py
│ │ ├── c_agent.py
│ │ ├── i_agent.py
│ │ └── r_agent.py
│ └── models/
│ ├── init.py
│ ├── agent.py
│ ├── client.py
│ ├── product.py
│ └── session.py
├── infra/
│ ├── init.py
│ └── database/
│ ├── connection.py
│ ├── migrations.py
│ ├── models.py
│ └── utils.py
├── requirements.txt
└── server.py
1 Like
I also have this issue. I even tho I’m making recently less requests it costs a lot more to me. I’m thinking about switching to claude code.
2 Likes
LITERALLY EVERYONES REQUESTS ARE LIKE THIS. Holy hell can you guys get your stuff together? Right now EVERYONE IS EXPERIENCING THIS. Every single prompt takes 50k+ tokens (I know this includes cache but STILL)
I will literally chargeback $200+ on my credit card if you guys don’t fix this - I have verbally and publicly supported Cursor in the past but this is absolutely batshit insane. PLEASE FIX.
8 Likes
Where is the bug? Every request uploads all your system prompts, rules, MCP server stuff, metadata about your project and recent files, etc etc etc. Your “test” even correctly shows you that it KNOWS that stuff - like your last commit etc - that is how AI works.
You give it context (aka tokens).
It produces the correct response.
There is no bug.
Bug? No bug. Just design.
They’re pushing over $100M+ in ARR, you really think they’d leave this to chance?
They’re not hiring interns to fumble prompts, they’re hiring pros to engineer how to drain your token usage efficiently.
It’s not broken. It’s working exactly how it’s meant to.
1 Like
Bug or not the 10x usage increase for less quality is still an issue.
2 Likes
@0xHACKS $100M ARR? - you’re a few months out of date - it’s $500MM ARR already. Not coincidental to this topic
@Xernive - everything is working properly - they gave us an “Opt Out” of the new billing exactly for this reason; they switched from per-call to per-token pricing, and if you didn’t opt-out or did not buy a years subscription, those days are over now. Anthropic charges Cursor per-token, so Cursor now charging us users likewise - they had to - their days of “buying customers with unlimited usage offers” have ended now. They won - real pricing begins now. There’s little point switching to anyone else as best I can tell - the pricing is the same everywhere - Anthropic charges $X, so providers who are not burning VC funds to buy users, must charge their users $X++… basic math… AI chews GPUs, GPUs chew electricity, electricity chews $, someone has to pay…
2 Likes