Currently, you’re charging based on token usage instead of requests — and frankly, this has been incredibly frustrating.
During code generation, the responses often stop or get cut off in the middle. Despite this, tokens are still consumed.
In my daily workflow, nearly 40% of my token usage is wasted due to incomplete responses.
Even when it says “Aborted, not charged”, I can clearly see that the token usage increases. If it’s truly “not charged”, then why does the token count go up?
Before releasing flashy features, please focus on fixing this fundamental issue:
Don’t cut responses midway.
If they are interrupted, provide an immediate “Continue” button.
Or at the very least, don’t count the tokens from incomplete/aborted outputs against our usage limits.
I think Cursor should open the code of their fork because I personally and I’m pretty sure most people are just tired of these stupid bugs like connection errors and so on. It’s just so stupid that in 2025 they can’t fix these obvious bugs (I have a very good connection around 300mbps and never encounter any problems in other apps).
I think the IDE developers should try to work with Code Agents. For example, with https://cursor.com/agents - their engineer could just drop a link to the bug before going to work and get a ready fix for arrival at the office! Only $60/month on the Pro+ tariff!
You have cited cases where an error leads to the fact that you are Not Charged for tokens. But if an error occurs inside a long process or the IDE thinks that the process has been completed successfully (after the first edit from Grok 4, yep), you will not see Not Charged.
