I am new to open ai and just want to try things. There are some scary numbers in the settings, that I cannot change, apparently.
I have set a 10$ “budget” under:
Project > Limits > Edit budget
So far so good. I did the same under:
Organization > Limits > Edit budget
However, when I scroll down that page a bit further, the shocker:
Usage limit
The maximum usage OpenAI allows for your organization each month.
$100,000.00
What??? I want to change that limit! No way I ever want to spend 100k! But apparently I cannot change the limit??
Now I feel very insecure about using the API at all, without lowering that limit.
And what exactly is a budget then, if not a limit? What are the terms different (“budget” VS “limit”)? I mean, my natural understanding of the words would be, that one can also be “over budget” and that would be super bad, if the actual true limit is 100k …
Please enlighten me about the definitions of the terms “budget” and “usage limit” under an organization. All I want is to set an absolute hard limit for ANY kind of usage, so that with maximum stupidity, I could only ever cause 10$ costs. How do I do that?
This topic is absolutely on point. It discusses valid trepidation.
There is no usage limit that will shut you down.
The only response is to not put more in prepaid credits than you can afford to lose (or less, because you can go into the negative for hours before shutoff);
The actual response is for OpenAI to put back hard limits, and further improve the hard limit optional features.
Budget must be renamed in the UI, as it presents a false premise, only sending emails.
A solution is not “get emailed”.
That is all anyone wants. OpenAI used to have that for an organization, with a straightforward presentation of a monthly limit after which the API would be cut off, with explanation:
The only way they suggest now when facing account abuse: go through the work of every project and to delete api keys, along with needing to find legacy user keys and delete those also, with the last not able to be deleted until replaced. Bad: every high-security application server and every user must be refreshed with new credentials.
Doing such would damage a whole bunch of setups and deployments of rights to users, and take time during continuing abuse.
There needs to be a master “off” switch for an organization, and the same on projects.
A project, being a stupid overloaded name for a rights sub-container in an authentication domain, makes you think that you could could give a certain amount of credit there dedicated to a project. But no: again, amount of resetting monthly usage to just get an email if exceeded.
There needs to be a lifetime spend limit on a project, a hard cap with tracking shown, that if set, immediately turns off usage until you increase the dollar figure further.
There needs to a more granular limit (that works) for recurring usage, that shuts off an application’s project for its day or week budget, detecting and limiting abuse over a maximum. The rate limiter already stops and breaks apps on many model’s requests per day rate limit; this should be the developer’s choice in dollars.
API key revocation shall collect a “reason” field, and report it in 403 API error messages (“you are fired”)
Thanks for taking the time to report your legitimate hesitation.
If I understand correctly, you are saying, that whatever I do, I will only be able to use models as much as my prepaid credits allow. Am I understanding it correctly? Or am I running the risk of my credit card being charged, if I overstep budget?
The system did not block further calls because I still had credit. As long as the balance shows money, the API keeps working. Rate limits do not touch the balance.
The tooling with “budget” is very limited, and it sucks, that I need to go there months by months and make a transaction every months, instead of paying once and set proper limits. Not that “budget” non-sense. “Budget” should be named “e-mail alerts” or something, but not “budget”.
