Sudden spike in consumption

API
1

I got upgraded to “level 4”, all of a sudden i get huge spike in consumption (at hours where no-one is using my api calls…)
i revoked all keys, reinstated new ones
tested with minimal token consumption (not even 200in /200 out) over 50 cents for just that call !! nowhere near where 4o-mini should be !!

also the models used are not those used by the API… very suspicious

Billing also bypassed the limit /budget that was instated, couple hundred dollars !!

Support doesnt reply…

Finally some support feed back (after 3 days…) and a refund - however, the consumption remains totally abnormal and doesnt match the models I use… i am waiting on support to get more info…

1 Like
2

Did you also reach user keys? https://platform.openai.com/settings/profile/api-keys

Did you see if other members have been added: https://platform.openai.com/settings/organization/members

If you have a compromised application from the design and keep rotating new keys into it, you’re only creating minor inconvenience to the abuser.

3

I dont think its an abuse - the keys do not seem compromised, i chnaged them for testing - the consumprion matches (timewise) the calls i have made after recrediting my account, so that doesnt come from elswhere.
It sounds more like a change in tier/ billing by OpenAI but they are not answering

4

The usage tier just allows you higher rate limits and higher maximum monthly spend. It should not impact costs, and it would be hard to imagine a mechanism where it would (unless similar system of “tier 7” could allow a partner negotiated secret discounts - were there such a thing).

I would look at the legacy usage dashboard, and see if the pattern for number of calls and token consumption, by models selected in the drop-down, and checking all the models, matches your pattern of use.

5

yes i checked legacy, thats how i figured some model that i do not use in my API calls have neen used
are you OPENAI ?

6

I’m just some random guy with advice, I cannot observe accounts.

You don’t use particular models but they have usage → someone else is using your models.

Pretty straightforward.

Usually something bad in practice, like hard-coding API keys in code, easily committed to repos or seen by others, or having any client software connect to OpenAI instead of to your own backend, allowing extraction and discovery of keys.

See, for example, this discovery of bad implementation: