I have already reached out to support regarding this issue but wanted to know if anyone else is experiencing this issue.
On our platform org we have 10s of projects, most with individual budget limits. We have always accepts these as hard limits as described in the documentation.
However today we have observed that the LIMIT value for all projects on our org level project page
is now displayed as “-” irrespective of if we have set a project level limit.
Having confirmed on a test project with a budget set to $1, I have been able to exceed the project budget limit of $1 for the project. A warning email is sent, but I can continue to make API calls without restriction. In addition, from a org reader account, and project owner, I have been able to modify the project level limit despite the highlight at the top of the page
You do not have permission to edit these settings
Only members of your organization with the “Owner” role can edit these settings.
This appears to be a critical issue, not being able to enforce individual projects budgets is a breaking issue.
Is anyone else experiencing this particular issue? Have I overlooked any recent changes that require updating to enforce project limits?
Thank you so much for taking the time to read.
3 Likes
Now confirmed with OpenAI support that project limits are no longer hard limits, and are notification only and that the documentation is incorrect it what it specifies.
Noting as well that “Monthly Budget Limit Silently removed” previously reported the silent change of org level monthly budget limits changed from hard limits to soft notification limits. I have been able to verified the org level budget change (it’s simple to setup a test project, but not a test org).
For anyone else in a similar position, the currently plan for our team is to implement a monitoring process, that will then restrict a project’s models when they hit the threshold. For now it looks like this is the only way to achieve the same effect as a hard limit. Any other suggestions would be welcome.
I am also aware of other possible solutions like using an LLM gateway such as LiteLLM which does include it’s own budget management. However this would not solve all our use cases, as I have yet to identify an LLM gateway/proxy that supports all OpenAI API endpoints (for example the realtime API over WebRTC and fine tuning). If anyone does have a robust solution that they could recommend, that would be really appreciated.
Well, this is not exactly a solution, but just in case you haven’t seen yet, there is the possibility to create an admin key to monitor your usage without depending just on email notifications.
https://platform.openai.com/docs/api-reference/administration
You can use this as and additional layer of monitoring, more integrated with your projects.
One more thing to notice, is that sometimes the usage data has a delay. Perhaps that is one of the reasons why the hard limit was disabled, as some delay in accounting for recent usage could still let requests go through.
2 Likes