Disappointment/problem with the credits usage

Discussions
1

I’m opening this discussion because the way Cursor calculates credit usage feels strange to me. I’ve already sent an email to support, but I wanted to share my experience.

The situation is as follows: I paid for a PRO plan to try Cursor, with the intention of upgrading to a PRO+ plan later if it convinced me. I used it for barely three days and all my credits were consumed.

I didn’t understand why they ran out so quickly. From the dashboard, I could see a very high level of usage, but I assumed it was normal. I enabled On-Demand mode and started monitoring the consumption more closely. At one point, I asked Cursor to perform a simple task: I had an error in a log file and wanted it to check where it was being generated and what the possible cause could be.

Cursor started iterating and repeatedly outputting the same messages. When I checked the dashboard, it had consumed approximately $1 in a very short amount of time, so I stopped it immediately. I didn’t know that in these situations credit usage could become almost infinite. And I find that extremely concerning, because the tool was clearly not doing what I asked—it was an AI failure. Why should I have to pay for an AI error?

Honestly, this feels very close to being a scam. Support responded with generic suggestions, such as:

  • Starting a new chat, since long conversations can accumulate confusing context

  • Explaining the task differently or breaking it into smaller steps

  • Trying a different model

  • Using Plan Mode first

My response was essentially: ok, thanks for the information, but it’s already too late. I didn’t know this. I would give it instructions and go do something else, and when I came back, sometimes it had been iterating endlessly and consuming credits the entire time.

Support has since replied again, saying they can see the issue in the usage breakdown and that they’re going to escalate it, but I don’t know what will come of that. The reality is that I’m now afraid to use Cursor, because I don’t have infinite money.

I’d like to know if anyone else has experienced something similar, because for me this is a real problem, and until it’s resolved, I don’t think I’ll use the tool again.

1 Like
7

There’s no doubt for me AI tools are very expensive these days.

I’m on the $200 one which for my current situation is on the painful side.

Composer-1 is my favorite model but it does churn up the tokens. As a rule of thumb if I’m doing donkey frontend work I use the Auto model, if its a big full stack feature covering a lot of the codebase I use Composer and Opus for any tricky bits.

I try to avoid using screenshots too and pressing Ctrl+N frequently

8

Yes, I too have had a similar issue easily using >1.5m tokens on the most trivial of prompts. People can make incredible sacrifices and work incredibly hard for money in life these days.
I hope you get somewhere with the ticket you raised. I will take it as a lesson learned.

9

There are a few takeaways when it comes to cost.

First, model choice matters. Each model has different speed, token price, and behavior.

Every model has pros and cons. There is no single model that fits every task.

Think of models like tools. A knife for cake is different from one for wood. You can cut wood with a sword. It works, but the cost is high.

Cursor pricing is based on tokens. More tokens mean higher cost. Some tools charge per request or per prompt. They still use tokens, but the cost is less visible.

Cursor is more transparent about usage. That makes spikes easier to notice.

For heavy agent work, Pro can be risky since its mainly for tab user. Pro+ or cheaper fast models fit better for heavy usage vibe code.

This matters even more during loops or retries. Costs can grow fast without limits.

10

Thanks, I agree there are many takeaways and I learnt many lessons. To be fair, I simply did not read enough, test or practice.
If you dont touch the big models you can have weeks of fun on $20.

11

Currently facing the exact same issue wasted a day waiting for a positive response. Will wait another day before taking the next step.