OpenAI API 500 and 503 errors: what they mean and what to do
Quick answer: Both errors come from OpenAI's side, not your code. A 500 means OpenAI's servers hit a problem processing your request; a 503 means the model doesn't have enough capacity right now. Retry after a short wait (following Retry-After on a 503), and check the live status below to see whether there's a wider outage.
The message you're seeing
500: The server had an error while processing your request503: Model temporarily overloaded
Is it down right now?
- OpenAI API · checked No problems
Live status, updated every minute. If it says “No problems”, the cause is most likely on your side: follow the steps below.
500 vs 503
- 500, “The server had an error while processing your request”: an issue on OpenAI's servers. OpenAI suggests retrying after a short wait, checking its status page, and contacting support if it continues.
- 503, “Model temporarily overloaded”: the model you asked for lacks capacity at the moment. Follow the Retry-After header when present, then retry.
What to do
- Retry with backoffWait, retry, and double the wait each time with a little randomness. Most 500s and 503s succeed on a later attempt.
- Follow Retry-AfterIf a 503 includes a Retry-After header, wait at least that long. It's OpenAI telling you when capacity is likely to be back.
- Try another modelA 503 is about capacity for a specific model. If your app can use a different model temporarily, it may keep working while the busy one recovers.
- Check for an incidentIf many requests fail for several minutes, look at the live status above and the API components on OpenAI's status page.
- Contact support if 500s persistIf one kind of request keeps failing with 500 while everything else works, report it to OpenAI support with the details of the request.
Is it an outage?
Occasional 500s and 503s are normal for any large API. If they become frequent across many requests, it's usually a wider incident. Our verdict above combines OpenAI's official status for its API components with our own check of api.openai.com every minute.
If you're getting 429 errors instead, that's about your rate limits, credits or spend limits: see our guide to OpenAI API error 429.
Questions
Is “The engine is currently overloaded” the same as a 503?
That wording appears in many older guides. OpenAI's current documentation lists the overloaded error as “Model temporarily overloaded” (503). The fix is the same: wait and retry.
Should I retry 500 errors automatically?
Yes, with backoff and a limit on the number of attempts. OpenAI's own advice for 500 errors is to retry after a brief wait.
Why does only one model return 503?
Capacity is managed per model, so a popular model can be busy while others are fine. Switching temporarily is a reasonable fallback.
Want to know when OpenAI API has a real outage? Follow our free outage alerts. We only post real outages, and again when they're fixed.
Related error guides
Based on official documentation, checked 18 October 2026: OpenAI: API error codes. Error messages and limits change over time, so the provider's own docs are the final word.