When ChatGPT spins forever, shows "A network error occurred," or the API starts returning errors mid-run, the useful question isn't "is it broken?" — it's which kind of broken. OpenAI can be fully down, degraded in one region, rate-limiting your account specifically, or perfectly healthy while your own network drops the connection. Each one has a different fix, and you can tell them apart in about 30 seconds.
ChatGPT is "down" in four different ways ¶
- Global outage — OpenAI's platform is down for everyone. Both chatgpt.com and the API fail worldwide.
- Partial / regional degradation — elevated errors or slow responses for some users; others are fine. Common during capacity spikes after a new model launch.
- Rate limit (HTTP 429) — the service is up, but it's refusing your requests because you've exceeded your quota or hit a burst limit. This looks like an outage but isn't one.
- Local issue — your network, VPN, a browser extension, or a corporate firewall is blocking the connection while OpenAI is healthy.
The 30-second check ¶
Run these in order — cheapest test first:
- Multi-region probe. Check chat.openai.com (or api.openai.com) from US East, US West, Europe, and Asia at once. If all four fail, it's a confirmed global outage. If some pass and some fail, it's regional. If all four pass, OpenAI is up and the problem is on your side.
- Read the publisher feed. Our OpenAI status page mirrors OpenAI's own incident feed, so you can see whether they've acknowledged an incident and which component (ChatGPT, the API, or the Playground) it affects.
- Check the AI hub. The AI provider status board shows OpenAI next to Anthropic, Google, Groq, and others — handy for telling "OpenAI is down" apart from "the whole AI layer is having a bad day."
Outage or rate limit? Read the error ¶
The single most misdiagnosed ChatGPT problem is a rate limit mistaken for an outage. The HTTP status code settles it:
- 429 Too Many Requests — you're being throttled. OpenAI is up; you've hit a requests-per-minute, tokens-per-minute, or daily quota. Back off and retry with exponential delay. What triggers a 429 and how to handle it →
- 500 / 502 / 503 — server-side failure on OpenAI's end. This is the real outage class. Nothing you change will fix it; wait it out.
- 401 / 403 — auth, not availability. A bad, revoked, or out-of-credit API key. The service is up.
The app and the API fail independently ¶
chatgpt.com (the consumer app) and api.openai.com (the developer API) run on shared infrastructure but have separate front doors, and OpenAI's status page reports them as distinct components. It's common for the API to be degraded while the web app is fine, or vice versa. If your app works but your integration is erroring, probe api.openai.com specifically rather than assuming the whole platform is down.
Symptom → cause → action ¶
| Symptom | Likely cause | What to do |
|---|---|---|
| All 4 regions fail | Global outage | Wait. Watch the OpenAI feed for an ETA. |
| Some regions fail | Regional / capacity | Retry, or route through a healthy region. |
| 429 in the API | Rate limit (up) | Exponential backoff; check your usage tier. |
| App works, API errors | API-only degradation | Probe api.openai.com directly. |
| All regions pass, only you fail | Local / network | Try another network, disable VPN/extensions. |
What to do while ChatGPT is actually down ¶
If the multi-region check confirms a global outage, there's no client-side fix — but you can (a) bookmark this page and re-check every few minutes, (b) fall back to another provider on the AI hub if your workflow allows it, and (c) if you're building on the API, make sure your code treats 5xx and 429 as retryable with backoff so a short blip doesn't take your own app down with it.
FAQ ¶
Is ChatGPT down for everyone or just me?
Run a multi-region check on chat.openai.com. If all four regions fail the same probe, it's down for everyone. If they all succeed, the outage is local to your network, browser, or DNS — not OpenAI. A single check from your own laptop can't tell those apart, which is why the regional pattern matters.
Why does ChatGPT say "network error" but the site loads?
That message usually means a long streaming response got cut off — the page loaded, but the connection holding the model's answer dropped. It's frequently local (flaky Wi-Fi, a VPN, or an aggressive extension) rather than an OpenAI outage. Retry once; if it persists across a different network, then check OpenAI's status.
Does a 429 mean OpenAI is down?
No. A 429 means the service is up and deliberately refusing your requests because you've exceeded a rate or quota limit. It's the opposite of an outage — the server is healthy enough to enforce limits. Back off and retry, and review your account's rate limits if it keeps happening.