Totally confused and honestly a bit frustrated here. Uploaded a ~65k token document just to see what it could do with the new April 2026 Deep Research Max preview models in AI Studio.
Because it’s a preview model, I played it safe and set a strict spend cap of €1 euro Initially. It failed after I checked back 30 minutes later, with no tokens spend, so I though I’d raise the spend cap to 8 euro.
Well… the model errored out again with no actual token calculation for billing after I check back 90 minutes later (according to the documentation it will take 20 to 60 minutes to come up with an answer) and I deleted that old chat.
Tried again with tweaking but same outcome. Apparently with some or at least the latest chat it got stuck in some crazy internal agentic loop for over 92 hours straight on the backend. Even though it already outputted that it failed and it showed no tokens to be paid for.
I turned off the api key for future chats and figured the model was not ready yet.
4 days later I get an email: Action Required: Your Gemini API services paused due to spend caps.
*forum only allows 1 embed for new users so had to combine the 4 images
For some of you maybe nothing crazy, but somehow I was 138 euro overdue on my 8 euro spending cap.
I assume somehow It just kept eating the input doc over and over again after which something must have changed and it billed it all at once suddenly 4 days later.
I dove into billing and it had processed over 26 MILLION input tokens and my spend cap completely failed to stop it.
I contacted billing support. After a ton of back and forth in chat (and even a Google Meet call with a rep just now), they agreed to a partial courtesy adjustment, but they are absolutely refusing to waive the remaining balance of about €18 (+ lost 8 euro of my balance that was already in my account)
Support keeps using the script that “API management is a shared responsibility” and that this is the absolute maximum they are allowed to refund in their system.
How is this shared responsibility?
-
I used the built-in spend cap. AI Studio completely ignored it and instead of a 10 minute delay, there a a 4 day delay.
-
The model hit an internal error, failed output, didn’t bill any tokens but still looped for 4 days straight.
I know €18+8 isn’t the end of the world, but this could have gone a lot worse with different limits. Why are developers forced to pay for a platform bug and a broken safety feature?
Has anyone else had the spend cap fail this badly?
