Since yesterday, 15th of oct, errored and/or aborted requests no longer show up as ‘Errored, not charged’ in the usage dashboard, instead appearing to be charged as normal. Same with all models.
Is this expected?
Steps to Reproduce
Start a request.
Abort it or disconnect internet / close Cursor to make it error.
Expected Behavior
Errored / aborted requests should not be charged, and show up as ‘errored, not charged’ in
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Yeah, I have the same problem. I keep trying to direct the neural network correctly, but if it makes a mistake, I undo it, and still nothing comes back. I thought maybe I could just undo the changes and then everything would be fine, but no. It doesn’t work.
Please do something about this. As soon as the model starts doing something wrong, I stop it and they charge me. Even if it hasn’t written anything yet, they still charge 2 cents. Undo all doesn’t help.
Same here, it’s happening to me too. Whenever the AI goes off track and I stop or abort the request, it still charges me every single time. It’s really frustrating because it never used to be like this before.
Same here. Whenever I see model is implementing a wrong code or not doing or going off the track , i stop it and earlier it didn’t charge me but now it is charging me for every single stops.
@deanrie This needs to be fixed asap with critical priority. This will be reason to switch as I’m losing my money.
Hey everyone. I just checked with the team, this isn’t a bug, it was an intentional change.
We started charging for partial requests since they can be used to bypass the system and still consume resources. We believe it’s fair to charge for runs that get stopped mid-flight.
This change also helps make billing more transparent and consistent, now any compute usage, even if a request is stopped early, is tracked accurately.
So you are saying if cursor goes off track or start implementing wrong code and if we stop it, we will still get charged. This happens many times. How this is fair when cursor makes mistake ? This is not fair at all.
Since it is not a bug and intentional, I will switch to Claude code next month because I’m going to lose a lot of money.
Thanks for the response. Unfortunately, along with the recent price updates, this makes Cursor many times more expensive than comparable tools. I’ve canceled my subscription.
Honestly, this would all make sense if Cursor’s accuracy was 100%. But when it’s not and the burden shifts back to the user to fix the AI’s mistakes it kinda defeats the whole purpose of using an “AI-powered” editor in the first place.
And now with the token-based charging, it’s getting ridiculously expensive. Like seriously, it costs more than my salary at this point . So yeah, thanks for the service, but I’ll probably switch to GitHub Copilot next month or just go back to old-school coding. At least that doesn’t charge me extra every time I breathe.
Also, one honest note ever since you guys implemented these new systems, the models have started hallucinating way more. I don’t know what happened, but it’s definitely not the same stable experience it used to be.
@SHEHROZFF@Avinaash_K the AI model accuracy is not 100% as it does not work like a mathematical calculator but with probability of text output matching your request. The AI models come from various AI providers like OpenAI, Google, Grok or Anthropic who are processing your AI requests and providing the response. We receive from them the token based consumption information which you can see in your Usage.
When someone interrupts a request that request was already submitted to the AI provider who does charge for the consumption.
. So yeah, thanks for the service, but I’ll probably switch to GitHub Copilot next month or just go back to old-school coding. At least that doesn’t charge me extra every time I breathe.