I’m writing after an incident. I’ll keep the story short.
One of my API keys was abused by a frustrated ex-developer of my company. Luckily, I was able to see in a short period of time that the credits of my account had been funded twice in less than 24h (normally, it is +/- once per month). I quickly stopped the bleeding. 108 millions tokens.
I then tried to find how to limit this risk.
It is not possible. I am on tier 5. OpenAi allows me 200 000$ per month of usage. Which I never plan to go that far, at least not before a few more years and a big business shift.
(our costs are around 200$ per month).
It is only possible to have notifications, to set a budget, but not a hard limit. At least, not for personal and business accounts. It is only possible for Edu and Enterprise accounts.
I cannot explain why, if the possibility to do it exists, why not give it to all accounts? The only reason I have in mind, gives me a dark opinion of OpenAI.
I’m considering leaving OpenAI for this unique reason.
Can someone find a good reason why they wouldn’t allow us to limit our consumption ? (or to set this 200 000$ to an amount we judge reasonable ?!).
I prefer to see the positive, and wouldn’t like to boycott this company.
Such feature used to exist, but my guess is it was disabled due to some operational complexity.
If you need a hard limit, you can use openrouter.ai as a proxy to your openai keys (BYOK). They have options to put a hard limit daily/monthly on costs.
The only restriction would be that since last time I checked, they only support chat completions.
I feel your pain. And it’s INCREDIBLY frustrating. Having set my “Budget” or “Pre-funding” to $10, I had foolishly assumed this WAS a hard limit. I set it to $10 BECAUSE I wanted a hard limit at $10, to avoid exactly your scenario.
Well, the other day I was coding an app and figured one part of the app might benefit from the strongest thinking model available. OF COURSE I looked up the price first.
GPT 5.2 Pro (high). $12 per million tokens. So each of my request/responses ought to cost around $0.30 to maybe $1.00 at most. That seems reasonable for my use case, so I plugged it in.
Of course I had an issue with the code, so I fixed it and tested again. Another issue, but progress. Fixed and tested again. And now I get an error, something about hitting some limit.
”A limit? Really? Have I reached $10 in only a few requests?”
I log into my OpenAPI dashboard and look at usage. Only used 18,000 tokens. Only 7 requests. Seems about right. But wait!!!
$82!!!
I asked Chat about what might be going on. I was informed that there are input tokens (I know) and output tokens (yes, and I’m aware of the costs). AND he informs me there are THINKING Tokens. So I must have blown through several million THINKING tokens.
If there can’t be a hard limit (which there should be), then AT LEAST make sure documents are COMPLETELY transparent, so we don’t use models we cannot afford to mess with.
There are also the previous chat history tokens that you send again as input, or are maintained and attached for you if using the Responses API with previous response ID or “conversations” API ID for stateful conversation.
Final cost of such an API call = $1.145739 + $2.16384 = $3.309579 (≈ $3.31)
OpenAI does tend to “cap” the account when “refresh/recharge” is disabled so payment card isn’t automatically charged, and shortly after your credit balance goes negative.
