My chat app is scheduled for review tomorrow, but it has been completely unavailable for about three hours. Every request to GPT 5.6 is returning a 429 error with the message "rate_limit_exceeded" and "temporarily unavailable." I haven't used the API much during the past 24 hours, and the app has no customers yet, so I don't think I'm hitting an account-level usage limit. I also can't find any public outage notice. Is anyone else seeing elevated error rates, or is there another place I should check?
2 Answers
A 429 paired with “temporarily unavailable” often indicates temporary provider capacity problems rather than your own usage. Check the limits for that exact model, since newer models can have different quotas, but I’d also add exponential-backoff retries and a fallback model before the review. That way, a brief outage results in a short delay instead of a completely broken screen.
You may not be imagining it. I’ve seen intermittent elevated error rates across several models, with some endpoints behaving differently from others. These incidents don’t always appear on the public status page right away, and the failures can come and go. A fallback provider or model, plus retries with a capped delay, would make the app more resilient while the service recovers.
It’s a chat app, and it has been unavailable for roughly three hours. This model had worked normally for the past few months, so the problem only started this morning.