Default organization does not take organization rate limits into account

My organization account has a set monthly budget and email notification thresholds that are defined in “Settings” > “Organization” > “Limits”.
This approach has prevented unexpected high costs in the past.

The “Projects” (see here for more info) feature that to the best of my knowledge has been introduced recently has made it possible to circumvent the Organization spending limits.

Everyone from my organization has been added to “Default project”. In this Default project it is not possible to set the limits because “You must be on an enterprise plan to set project spending limits”. As a result, the monthly API costs have exceeded our thresholds by a lot.

From a company admin perspective it seems incredibly misleading that the Organization usage limits can be circumvented like this, we had the correct safeguarding tools in place and all of sudden these don’t work anymore.

To prevent this in the future, how can we upgrade to an enterprise plan? We have an organization account, we thought we already were on an enterprise plan? Searching for “OpenAI enterprise plan” gives back results for ChatGPT for enterprise, which is not related to the current problem.

As a long time user of the platform this sequence of events that resulted in high costs has left a very sour taste. I’m curious whether the above also applies to other users and if so, this post can help raise awareness for this issue.

2 Likes

We have had this as well, but I am not sure this is related to the projects that were introduced recently. In our case the hard limit was set at 400$ and it allowed an overshoot to 820$!

2 Likes

Yes, it is absolutely appalling that there have been dozens of reports just here of people blasting right past their limits, to the tune of $1000 in billings or account balances going hundreds into the negative before any shutoff, and not a peep from the multiple OpenAI staff that frequent the forum to escalate this concern, and complete lack of safety from continued hacking campaigns against OpenAI users and credentials.

Or why it was even considered, when, apparently all prepaid accounts in March had a “prepaid accounts can’t set any limits” plastered over where the limits used to be, resulting in unalterable limits and further interruptions in service. To solve that, they disabled the limits completely, and then amid protest, only put a fake box back?

2 Likes

The intended behavior is that you should be notified by email when you exceed the soft limit, and the API should be disabled when you exceed the hard limit.
However, as discussed here, there are many reports that the hard limit is not functioning properly.