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

Your production application just threw a cascade of 500 errors, your monitoring dashboard lit up red, and users are flooding your support channels. Before you spiral into emergency mode, spend 30 seconds checking if this

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

Your production application just threw a cascade of 500 errors, your monitoring dashboard lit up red, and users are flooding your support channels. Before you spiral into emergency mode, spend 30 seconds checking if this is actually an OpenAI API outage or something on your end — the fastest way is our live OpenAI status monitor, which updates every 60 seconds from OpenAI's official feed.

How to check if OpenAI is actually down

Start with our live OpenAI status page. It pulls directly from OpenAI's infrastructure monitoring and breaks down every component separately — ChatGPT web, the completions API, embeddings, fine-tuning, Realtime API, and more. If you see "Operational" across the board but you're still getting errors, the problem isn't OpenAI's infrastructure.

Head to status.openai.com for OpenAI's official incident reports. When there's a genuine outage, OpenAI posts updates here within minutes — you'll see the affected services, error rates, and estimated recovery times. During the December 2024 outages, for example, OpenAI reported elevated API error rates and posted updates every 15-30 minutes.

Check component-level status specifically. The API has separate systems for different model families and endpoints. You might see GPT-4 running normally while GPT-4o throws errors, or the completions endpoint working while embeddings time out. Look at the exact service you're calling.

Third-party monitoring sites like DownDetector show user reports clustering around specific times, but treat these as confirmation rather than primary evidence. A spike in reports usually follows an outage by 5-10 minutes as users notice and post.

Is it OpenAI or is it you

Run these checks in order before assuming it's an outage:

  • Test with curl directly. Strip out your application code and hit the API with a basic request: `curl https://api.openai.com/v1/chat/completions -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"model":"gpt-4","messages":[{"role":"user","content":"test"}]}'`. If this works, your application layer has the issue.
  • Check your API key and quota. Log into platform.openai.com/usage and verify you haven't hit your rate limit or spending cap. Error 429 means rate limiting — that's on your end, not an outage. See our guide on OpenAI API error 429 and other API errors for the full breakdown.
  • Verify your network path. Try the same request from a different network. Corporate firewalls, VPNs, and DNS issues can block API calls. Some organizations block api.openai.com at the network level.
  • Test different models. If gpt-4 times out, try gpt-3.5-turbo. Model-specific issues happen — they're not full outages but degraded service for particular endpoints.
  • Check your error codes precisely. A 500 or 503 from OpenAI's servers suggests their infrastructure. A timeout with no response code points to network issues. Connection refused means routing problems, likely on your end.

What to do during an outage

When you've confirmed it's actually an OpenAI outage, here's your playbook:

  • Implement exponential backoff immediately. Don't hammer the API with retries — you'll make recovery slower for everyone. Start with a 1-second delay, then 2, 4, 8, maxing out at 60 seconds. Add jitter (random variation) to avoid thundering herd effects when service resumes.
  • Queue requests instead of dropping them. If your application allows it, store failed requests in a queue and process them when the API recovers. Users get delayed responses instead of complete failures.
  • Fail gracefully in your UI. Show users a clear message: "OpenAI's API is experiencing issues. Your request is queued and will process when service resumes." Don't let error messages bubble up raw.
  • Set a retry budget per request. Don't retry indefinitely — after 3-5 attempts over 5 minutes, give up and notify the user. Log these for post-mortem analysis.
  • Monitor OpenAI's status page for updates. They'll post ETAs once they've diagnosed the issue. Don't waste time contacting OpenAI support during an active outage — they're focused on recovery, not individual tickets.

Questions people actually ask

Q: Is ChatGPT down right now?

A: Check our live OpenAI status monitor — it shows separate status for the ChatGPT web interface and the API. They can fail independently. For broader ChatGPT issues, see ChatGPT not working — troubleshooting.

Q: Do I get refunded for API downtime?

A: OpenAI doesn't automatically refund for outages, but they've issued service credits after major incidents. You'll need to contact support through platform.openai.com if you want compensation for extended downtime.

Q: How long do API outages usually last?

A: Most resolve within 30-90 minutes. Major incidents can run 3-6 hours. Check status.openai.com for real-time estimates.

Q: Should I switch API keys during an outage?

A: No. This won't help and wastes time. Outages affect the infrastructure, not individual keys.

What to remember

  • Our live status monitor updates every 60 seconds — bookmark it for instant confirmation
  • Exponential backoff with jitter prevents you from making outages worse
  • Queue requests instead of dropping them when possible
  • Check component-level status — the API has many independent services
  • Don't contact support during active outages unless you're facing billing issues

---

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

Related help