API requests failing despite being well under quota

We are currently seeing errors on all our API calls despite being well under the quota and having very minimal simultaneous usage. Calls to gpt 3.5 result in a “exceeded current quota” failure, and calls to gpt4 result in “the model gpt-4-turbo-preview”. We’ve been facing these issues since yesterday afternoon, and our users are unable to use our feature as a result. We don’t see errors with the same models on playground.

I don’t see a failure on our status page, would love some guidance on what is going on. Our feature is fully down and blocked until we are able to access the api. Thank you!

3 Likes

This is directly impacting our business—our users aren’t able to use our AI features right now!

1 Like

While not a direct solution to your current issue, I would suggest eliminating as many single points of failure as possible.

It would probably be a good idea to set things up so you can fail over to Azure if you have a problem connecting to the OpenAI API.

My app (quanta dot wiki, open source) connects to OpenAI, Perplexity, Anthropic, and Gemini, so that users can never be disabled completely from using AI unless all four of those cloud services go down simultaneously. So it’s not that hard to accomplish this level of failover/redundancy because so many of these cloud services use nearly identical APIs.

If you had the time to write up a short tutorial or something sharing some pain points you encountered and solutions you’ve found to them, I’d be happy to pin it for visibility.

Thanks for the suggestions! For posterity, we were able to resolve this issue by updating our API key to a fresh one. Ideally, we’d be notified when active API keys are about to become stale due to whatever reason; in our case it was that the original creator of this key’s account became invalid. Thanks

To the best of my knowledge API keys do not become stale. They can become revoked if OpenAi detects they have been compromised, but they don’t expire.

If your key had been revoked you would have received an invalid API key error when using the key.

It’s possible you switching to a new API key and the resolution of the issue were entirely coincidental.