On Sep 10 I’ve spent the entire “Other Models API” and that was ok: the chat with gpt-5.6-sol-high 1M just stopped because there was not enough budget allocated (100% used).
But the next day was my renewal day and I noticed it was spent by 16% already. I checked the chat - and there something happened like icon was green and then switched to activity when I clicked on the chat name, although the chat were at the point where it stopped the last day.
Then I noticed for some reason the button on the chat textbox switched from “Microphone” to “Stop” - so I clicked it immediately to stop the action to prevent any additional unwanted budget spending. I never clicked on “Continue Working” button Then I noticed the subagents are still working - so I went there and tried to stop them individually and use Stop All button with confirmation after that. No actions helped - the agents were still active for some reason… So I closed the Cursor IDE and started it back - after that the chat was silent and no signs of weird activity.
Then in the dashboard usage I found the next lines:
Sep 12, 02:58 AM Free gpt-5.6-sol-high - Free
...
Sep 11, 08:09 PM Included gpt-5.6-sol-high 364.3K Included
...
I’ve contacted support - but found only uncooperative AI agent that tried to deflect the issues I’m describing and unwillingness to help. It was just terrible experience, so I hope someone here will be able to actually help me with this issue.
Steps to Reproduce
Have no idea, but I suppose the next steps could help:
Spend everything in the budget on gpt-5.6-sol-high 1M Other Models API
Chat will became stopped
Live till the renewal period not restarting the cursor IDE when your utilization became 0%
Check back the chat and confirm nothing happens there
In ~6 hours of other chats usage on “Cursor Models” check the utilization
Find out it for some reason not 0% anymore
Visit the dashboard usage and find there are some use
Expected Behavior
The chat should not suddenly try to auto-continue - the sudden stop due to no budget could break the context and spend user money for unwanted activity. Any continuation should be confirmed with the user with clear identification of what kind of issues it could have.
Hey, thanks for the detailed report, and sorry you had a bad support experience before.
I looked into what happened in this chat. On September 10, the agent started two sub-agents. They finished about 20 minutes before the chat hit the Other Models usage limit, and their completion notifications were still queued for delivery at that time. Cursor automatically wakes a chat when a background task finishes, and it did that as soon as your usage refreshed. That single wake-up call on September 11 at 8:09 PM is the 364.3K tokens, about 16% of your pool, that you saw. After that, it ran ./test.sh in the terminal, which didn’t finish until you clicked Stop. The Sep 12, 02:58 AM line was a retry to deliver those same notifications that you stopped after a couple seconds, and it cost $0.
You’re right. A chat stopped due to the usage limit shouldn’t resume on its own without confirmation, and Stop should clear those pending notifications instead of leaving them in the queue. I’ve passed both points to the team.
You can keep using this chat normally. Restarting Cursor cleared the queue, so it won’t resume by itself anymore. Until this is changed, if a chat ever stops due to the usage limit while sub-agents or a background command are still running, restart Cursor, or archive this chat, before the billing period refreshes, and the pending wake-up will be reset.
About a refund for this charge ($3.66, about 16%), please email [email protected] and the team can review your specific account. If anything else comes up in this chat, let me know here.
Thank you @deanrie , this helps! I know I ask too much - but is it possible to tune the AI on email support? I know it’s business decision, but it almost made me rage-quit the subscription (ticket T-F73253)… Surely it saves tons of costs and quickly deflect any request, but could it be more compassionate at least to the bugreports and take them seriously, maybe to recommend to create a ticket here? That was a challenge to get here with 0 knowledge…
About email support, that feedback is fair, and I’ll pass it on to the team that owns the support experience. Bug reports really should be handled seriously, not brushed off, and the suggestion to start a forum thread in cases like this is a good idea.
Thanks for sticking with it and making it this far anyway. If anything else comes up related to this chat, just reply here.
And the result of writing to [email protected] is the same: nothing could be done, according to helpful Sam AI:
Hi Sergei! Thanks for reaching out! Your ticket number is T-F74581 for your reference.
I’m not able to restore or credit that Other Models included usage back onto your monthly limit, and a teammate can’t reverse it either. Each model run incurs real provider cost, so usage that already ran stays on the account for the cycle. Your included usage refills on its own at the start of your next billing period on Oct 11.
When background or parallel subagents keep going after a normal Stop, use Stop All in the subagent tray. You can also keep on-demand off or set a low spend limit at .
You can review the full per-request breakdown under All Events here:
Let me know if you need anything else!
And I understand it can’t do anything, because it’s in sandbox, but is there people to do something with that?
Got it. It’s frustrating that the email auto-reply didn’t resolve your request, especially since the charge came from behavior you didn’t start yourself.
I can see the charge as a single wake-up call and I understand what caused it. For the details on that specific charge, let’s move this out of the public thread. I’ll pass it directly to the team (T-F74581) so a person reviews the case, not an auto-responder. I can’t promise an outcome in advance, but I’ll make sure it doesn’t get stuck.
On the behavior itself (auto-resume after hitting the usage limit and Stop not clearing the notification queue), I’ve already sent it to the team and it’s being tracked. Thanks for following through to the end.
Hey, quick update: the specific charge has been handled directly through your support ticket T-F74581. Check that thread for the details. The behavior side, auto-resume after a usage-limit stop and Stop not clearing the queued notifications, is with the team. If anything else comes up in this chat, just reply here.
