I believe there is a new bug in platform.openai.com . Since today, I do not see a spending budget button for both my organization and the projects within it. For the organization, I no longer see the spending cap I had put in place before. For the projects, I do see the caps, but I believe they are not being enforced. I am an organization owner, so it is not a permission issue, and I tried incognito mode for 3 different browsers, so it’s not a cookie issue either. I noticed that 2 days ago already a project went 2x over budget (1800$ on a 1000$ cap) while still being green which tells me that the whole spending cap subsystem might be completely broken right now. This is super important for us to keep our costs manageable, so we will appreciate quick help.
There have not been any functioning “hard limit” caps for about a year.
OpenAI removed them.
That is not true - up until 5 days ago they were functioning just fine - yes sometimes they will stop the usage few dollars above the max set but they were working. Now I cannot manage them at all
Granted: I see a revision in the limits UI for project and for organization, where we no longer see “set a budget” - which would send merely a notification email and was not the removed “hard limit”. Instead there is language for “Alerts” now, and for configuring all as only one type instead of appearing as a “budget” with more “email alerts” possible.
That is a better description of what was offered.
If you actually got a project turned off while you still had an organization balance, that is quite contrary to expectations and what has been happening.
You can visit this topic I created months back, where at least the current presentation in the platform is a good response, in that an API developer is not misinformed by the language.
I again push back on what you are saying - literally 5 days ago, a limit on a project will stop the project from being able to use credits. This was enforced loosely (something like eventual consistency, where a few minutes after the spending went over, it would only then prevent future spending), but it did. Now this functionality is esentially completely gone. I am saying this because i had multiple projects that triggered this 5 days ago
I am on your side, and that both projects and organizations must have hard and immediately working budget limits that can be set.
Google has working AI budget limits, on top of them offering post-paid billing and not expiring credits you are forced to buy in anticipation of future use, for future models and their pricing, and even dubious future persistence of your API organization not being arbitrarily banned and having those credits stolen. The frogs here are not just boiling, they are already cooked.
Welcome to the forum!
From what I can see, there are usage/rate limits and there are also billing/spend alerts. But the current wording around spend/budget controls seems to be about alerts/monitoring, not a guaranteed hard stop. (As already stated in this thread)
So I think what’s already been pointed out is accurate for the currently documented/current UI behavior: these budget/spend settings should not be relied on as hard caps.
That said, if your projects were actually being stopped before when they crossed the limit and that behavior changed recently, then Support/OpenAI staff would probably need to clarify whether that previous behavior was intended, changed or broken.
They are probably the best option now to help your organization with this.
Just noticed this feature was removed as well.
This was definitely something that worked, we could set limits and when hit would usually stop a little above the limit and error any requests that would be sent after or were waiting to be processed (batch request).
The removal of this feels strange, it was a nice peace of mind ensuring rogue code wouldn’t blow the budget, as well as an added security feature if an API key ever got leaked somehow. I’m not sure what removing something that worked perfectly fine achieves.
If anything I would have liked it extended so we could set per-API key limits instead of just per-project limits.
Welcome to the forum!
And thanks for adding this. That matches what OP described too.
Like stated, the current wording present these more as alerts now. OpenAI staff/support would probably need to clarify whether that change was intended.
Thanks for the response.
From what I can tell the only way now to manage limits is to set a monthly limit in the auto-recharge section of the billing page. It does mean there’s a reduced functionality of project level limits though, but it’s still something at the org level.
The question I have now though, if these old limits are removed, why does the UI still show the “x used of y limit” progress bar, when the y limit is no longer configurable. My project limit also still shows that previous limit number I had set at the end of the progress bar. This number seemingly can’t be changed any more, editing the alert limits also doesn’t seem to change it. Is there still a limit? When I reach it will requests error like before, or is this now just a UI artefact that essentially means nothing?
Here are two docs that may be useful for you to check (if you already haven’t):
Project / budget docs:
https://help.openai.com/en/articles/9186755-managing-your-work-in-the-api-platform-with-projects
Prepaid billing / auto-recharge docs:
https://help.openai.com/en/articles/8264644-how-can-i-set-up-prepaid-billing
I’d also still recommend contacting OpenAI Support for clarification, since they would be able to give more accurate information about your specific account/project behavior.
Small note: I’m not OpenAI staff or OpenAI Support, just another user here trying to help. 