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

When your production application starts throwing 500 errors or timing out on every OpenAI API call, you need to know within 30 seconds whether it's an actual outage or something you can fix. Check the live OpenAI status on this site — it updates every 60 seconds from OpenAI's official feed and shows component-level status for the API endpoints you're actually using.

How to check if OpenAI is actually down

The live status monitor on this site pulls directly from OpenAI's status feed and breaks down service health by component — chat completions, embeddings, assistants API, images, audio, and moderation endpoints. You'll see "operational", "degraded performance", "partial outage" or "major outage" for each. This updates automatically every minute, so refresh it if you're troubleshooting in real time.

Head to status.openai.com for OpenAI's official incident reports. During an outage, they post updates every 15-30 minutes with specific affected endpoints, error rates, and estimated resolution times. If you see "investigating" or "identified" status, it's confirmed — they know about it and you're not alone.

Check which specific endpoint is failing. A chat completions outage (gpt-4, gpt-3.5-turbo) doesn't necessarily mean embeddings or DALL-E are down. Look at the component breakdown. If you're getting error 500 or 503 responses, that's almost always server-side. Error 429 during normal usage volumes also suggests capacity issues on OpenAI's end, not a rate limit you hit.

Is it OpenAI or is it you

Before assuming it's an outage, isolate your setup. Run a test API call from a completely different environment — try the official API playground at platform.openai.com/playground. If that works but your application doesn't, the problem is local.

Check your API key validity at platform.openai.com/api-keys. An expired or revoked key gives you error 401, not downtime. Verify you're using the correct base URL: `https://api.openai.com/v1` — typos or old URLs cause mysterious failures.

Test from a different network. Corporate firewalls, VPN restrictions, and ISP issues can block OpenAI's servers. Use your phone's hotspot or a different WiFi network. If it suddenly works, your network is filtering the traffic.

Review recent code changes. If you pushed an update in the last 24 hours and requests started failing, roll back and test. The timing isn't coincidence — you may have introduced a malformed request body, wrong model name (gpt-4-turbo vs gpt-4-turbo-preview), or authentication header issue.

Check your rate limits at platform.openai.com/account/limits. If you're on a free tier or recently upgraded, you have specific requests-per-minute caps. Hitting those gives you error 429 with `rate_limit_exceeded` — see our guide on OpenAI API error 429 and other API errors for the difference between rate limits and actual outages.

What to do during an outage

Implement exponential backoff immediately. When you get a 500 or 503 error, don't retry instantly. Wait 1 second, then 2, then 4, then 8 — double the delay each time up to a maximum of 60 seconds. This prevents you from hammering their servers when they're already struggling.

Queue failed requests instead of dropping them. Store the user's input, show a "processing delayed" message, and retry automatically when the API recovers. Users accept a 2-minute delay far better than "your request failed, try again."

Set a retry budget — maximum 5 attempts per request. After that, fail gracefully with a clear error message: "OpenAI's API is temporarily unavailable. We've saved your request and will process it automatically when service resumes."

Switch to a fallback model if you have one configured. If gpt-4 is down but gpt-3.5-turbo is operational (check the component status), automatically downgrade for non-critical requests. This keeps your application functional during partial outages.

Do NOT delete your API key and create a new one. Do NOT cancel and re-add your payment method. Do NOT submit multiple support tickets. None of this helps during an outage — you're just creating more work when service restores.

Questions people actually ask

Q: Is ChatGPT down right now?

A: Check the live OpenAI status — it shows real-time status for both ChatGPT and the API, updated every 60 seconds. Green means operational, yellow or red means confirmed issues.

Q: How long do OpenAI API outages usually last?

A: Most degraded performance incidents resolve within 15-45 minutes. Major outages typically take 1-3 hours. Check status.openai.com for live updates — they post ETAs when they have them.

Q: Will OpenAI refund me for API downtime?

A: OpenAI doesn't publicly guarantee refunds for outages. If you have enterprise support, contact them directly through your account team. For standard API users, you're not charged for failed requests that return 500/503 errors.

Q: Should I switch away from OpenAI after an outage?

A: Every API provider has outages — AWS, Google Cloud, Azure all have public incident histories. Evaluate based on frequency and communication quality, not single incidents. If you need higher reliability, implement multi-provider fallbacks in your architecture.

What to remember

Check the live status monitor first — it updates every 60 seconds and shows component-level status
Implement exponential backoff with a retry budget of 5 attempts maximum
Queue failed requests instead of dropping them — users accept delays better than errors
Test from different networks and the official playground to isolate local problems
Don't create new API keys, cancel payments, or spam support during an outage — none of that helps

Related help

---

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