OpenAI API outage — what to do when the API is down

Your production app just threw a wall of 503 errors. Your users are complaining. You're wondering if it's your code or if OpenAI's infrastructure is actually having problems. Here's how you can tell within 30 seconds: check the live OpenAI status on this site (updates every 60 seconds from OpenAI's official feed) and look at status.openai.com for any red indicators next to "API".

How to check if OpenAI is actually down

Start with our live OpenAI status monitor. It pulls directly from OpenAI's status feed and refreshes every minute, so you'll see component-level health for the API, ChatGPT, and other services. If the API shows "Operational", the problem might be elsewhere.

Next, go to status.openai.com and look for specific components. OpenAI breaks down their status by service: "API" covers /v1/chat/completions, /v1/embeddings, /v1/audio and other endpoints. If you see "Degraded Performance" or "Partial Outage", that's confirmation. Click through to the incident details — OpenAI usually posts timestamps, affected regions, and which models are impacted (GPT-4, GPT-3.5-turbo, text-embedding-ada-002, etc).

Check the error code your requests are returning. During an actual outage you'll typically see HTTP 503 (Service Unavailable), 529 (Overloaded), or timeouts. If you're getting 429 errors, that's a rate limit issue on your end, not an outage — see our guide on OpenAI API error 429 and other API errors. Error 500 can be either your request format or a backend problem, so cross-reference with the status page.

Is it OpenAI or is it you

Before assuming it's an outage, rule out your own setup. Test with curl directly from your terminal:

```

curl https://api.openai.com/v1/chat/completions \

-H "Authorization: Bearer YOUR_API_KEY" \

-H "Content-Type: application/json" \

-d '{"model":"gpt-3.5-turbo","messages":[{"role":"user","content":"test"}]}'

```

If curl fails with the same error, it's not your application code. If curl works but your app doesn't, debug your HTTP client library, headers, or payload formatting.

Check your network path. Corporate firewalls, VPNs, and cloud provider egress rules can block api.openai.com. Try the same request from a different network or a clean EC2 instance. Verify your DNS is resolving api.openai.com correctly — run `nslookup api.openai.com` to confirm.

Look at your API key status at platform.openai.com/api-keys. An expired credit card or usage limit can cause sudden failures that look like outages but aren't. Check platform.openai.com/usage to see if you hit a spending cap.

What to do during an outage

1. Implement exponential backoff immediately. If you're not already doing this, add it now. Start with a 1-second wait after the first failure, then 2, 4, 8, up to a maximum like 60 seconds. Don't retry on 4xx errors (those won't fix themselves), but do retry on 5xx and timeouts.

2. Queue your requests instead of dropping them. Use Redis, RabbitMQ, or your cloud provider's queue service to store failed requests. Process them when the API comes back. Your users see "processing" instead of hard failures.

3. Switch to a degraded mode if you have one. Serve cached responses, show static content, or temporarily disable AI-powered features. Don't let your entire application go down because one external dependency failed.

4. Monitor OpenAI's status page actively. Set up monitoring on status.openai.com's RSS feed or use our live status page. You'll get updates as OpenAI identifies, investigates, and resolves the issue.

5. Don't spam the API with retries. Hammering the endpoints during an outage makes recovery slower for everyone. Cap your retry attempts to 3-5 per request.

What NOT to do: don't delete your API keys and create new ones (doesn't help), don't flood support tickets about known incidents, and don't switch to a different model thinking it'll bypass the issue — during infrastructure outages, all models are typically affected.

Questions people actually ask

Q: Is the OpenAI API down right now?

A: Check the live OpenAI status on this page or status.openai.com. If the API component shows anything other than "Operational", there's a confirmed issue. Our status monitor updates every 60 seconds.

Q: How long do API outages usually last?

A: Most are resolved in 15-45 minutes. Major incidents can take 2-4 hours. OpenAI posts updates on status.openai.com as they investigate. The longest outages in 2024 were around 6 hours, but those are rare.

Q: Will I get credits for downtime?

A: OpenAI doesn't automatically issue credits for outages. If the outage significantly impacted your business, contact them at platform.openai.com/account and document the specific impact. They've issued credits in the past for extended incidents, but it's case-by-case.

Q: Should I have a backup API provider?

A: For critical production applications, yes. Consider having fallback logic to Anthropic, Google, or Azure OpenAI Service. It adds complexity, but eliminates single-point-of-failure risk.

What to remember

Check our live status monitor first — it updates every 60 seconds from OpenAI's official feed
Test with curl to isolate whether it's your code or the API
Implement exponential backoff and request queuing before the next outage hits
Don't retry 4xx errors, but do retry 5xx and timeouts up to 3-5 times
Have a degraded-mode plan so your app doesn't completely fail when OpenAI does

Related help

---

*openai-support.com is an independent resource, not affiliated with OpenAI Inc.*