Refresh & start a new chat
A reload or a fresh conversation clears many transient 400s and stuck/over-long chats.
ChatGPT errors, explained
Saw a number like 400, 401, 403, 429, 500, 502, 503 or 1020? Each one tells you something about where the problem is. As a rough guide, 4xx codes point to your side (a bad request, login/auth, permissions, or a rate limit), 5xx codes are OpenAI's servers, and Cloudflare's 1020 is a network/access block. This independent page breaks each one down — and shows how to keep working while OpenAI fixes the rest.
The full table
These follow standard HTTP status codes (plus Cloudflare's 1020). Use the "side" column to decide whether to wait it out or fix something locally. Codes and behaviour can change — verify on OpenAI's pages.
| Code | Meaning | Whose side | What to do |
|---|---|---|---|
| 400 | Bad request — the request was malformed or couldn't be processed as sent | You | Refresh, start a new chat, clear cache/cookies, try incognito. On the API, check your request body and parameters. |
| 401 | Unauthorized — authentication failed (login session or API key not accepted) | You | Sign out and back in, clear cookies, try incognito. On the API, confirm your key is valid and sent correctly. See login problems. |
| 403 | Forbidden — access to the resource isn't permitted (region or permission) | You | Disable VPN/proxy, switch network, use a supported region, confirm your account has access. |
| 429 | Too many requests — you hit a rate limit by sending too many messages quickly | Rate limit | Slow down, close extra tabs/devices, wait a few minutes. On the API, reduce frequency + use backoff. See the 429 guide. |
| 500 | Internal server error — a problem on OpenAI's side | OpenAI | Can't fix locally. Wait, refresh after a pause, check status.openai.com. |
| 502 | Bad gateway — an upstream server error on OpenAI's side | OpenAI | Wait and retry after a pause; don't retry aggressively. Check the status page. |
| 503 | Service unavailable / "at capacity" — servers overloaded or under maintenance | OpenAI | Wait, refresh after a pause, check status. See the at capacity / 503 guide. |
| 1020 | Cloudflare access denied — block tied to your IP, VPN/proxy or network | You / network | Disable VPN/proxy, clear cookies, switch network, update browser, fix date/time. See the 1020 fix. |
Don't just sit there
If you're seeing a 500, 502, 503 or "at capacity", the fix is on OpenAI's side — but you don't have to stop. A multi-model workspace lets you switch to another AI and finish what you were doing, then go back to ChatGPT once it recovers.
When ChatGPT throws a 5xx or hits capacity, switch to Claude, Gemini or another model in the same place and keep going for free — no waiting for OpenAI to recover. Compare answers across models too.
OfficialThe first place to check: live incidents, outages and maintenance.
OfficialOpenAI's own troubleshooting and error articles.
GuideHow to confirm an outage and what to do.
Fix the ones on your side
If the status page is green, a 400, 401, 403 or 1020 is usually local. Work through these in order — they clear the large majority of your-side error codes.
A reload or a fresh conversation clears many transient 400s and stuck/over-long chats.
This rules out a bad cache, cookies and browser extensions in one step — a common cause of 400/401 errors.
A blocked IP, VPN or proxy is the usual cause of 1020 and some 403 access errors. Try mobile data or another Wi-Fi.
Clear ChatGPT cookies and cache, update your browser or app, and confirm your device date/time are correct.
For a 401, sign out and back in. On the API, confirm your key is valid and sent correctly. See login problems.
For a 429, wait a few minutes and stop rapid retries. Close extra tabs or devices on the same account.
Per-error guides: something went wrong · network error · too many requests · error 1020 · at capacity / 503 · login problems
All error guides
FAQ
Short answers on what each code means, which side it's on, and how to fix the ones you can. Always verify current behaviour on OpenAI's official pages.
ChatGPT error codes mostly follow standard HTTP status codes. Roughly, 4xx codes (400, 401, 403, 429) point to something on your side — a bad request, an auth problem, a permission/region block, or a rate limit. 5xx codes (500, 502, 503) are OpenAI-side server problems. Cloudflare's 1020 is a network/access block. The fastest way to know which side you're on is to check status.openai.com first.
Check status.openai.com. If OpenAI reports an incident, the error is on their side — wait and retry. If status looks green, the cause is usually local: try a refresh, an incognito window, a different network, or disable your VPN and extensions. As a rule of thumb, 5xx codes and "at capacity" are OpenAI-side, while 4xx codes and 1020 are usually your-side.
A 400 "bad request" generally means the request sent to ChatGPT was malformed or couldn't be processed as sent. In the app this can follow a corrupted session, a stuck conversation, or bad cached data. Try refreshing, starting a new chat, clearing cache/cookies, or an incognito window. On the API, check your request body and parameters.
A 401 "unauthorized" is an authentication problem — your session, login, or API key wasn't accepted. In the web app, sign out and back in, clear cookies, or try incognito. On the API, confirm your API key is valid, active, and sent correctly. Always check status.openai.com in case there's an auth-related incident.
A 403 "forbidden" means access to the resource is not permitted. This can relate to region or permission restrictions, or a blocked request. Try disabling a VPN/proxy or switching network, make sure you're using a supported region, and verify your account has access to what you're requesting. If it persists after the standard fixes, check OpenAI's help pages.
A 429 "too many requests" means you hit a rate limit by sending too many messages in a short time. Wait a few minutes, slow down, and close extra ChatGPT tabs or devices using the same account. On the API, reduce request frequency and use exponential backoff. See the dedicated 429 guide for details.
These are OpenAI-side server errors: 500 is an internal server error, 502 is a bad gateway, and 503 is service unavailable (often shown as "at capacity"). They usually mean the servers are temporarily overloaded or under maintenance. You can't fix them locally — wait, refresh after a pause, and check the status page. Aggressive retrying can make things worse.
Error 1020 is a Cloudflare "access denied" block tied to your IP, VPN/proxy, or network — not a standard OpenAI status code. Fixes are on your side: disable a VPN or proxy, clear ChatGPT cookies, switch network or device, update your browser, and make sure your device date/time are correct. See the dedicated 1020 guide.
Usually no. Most error codes are temporary server issues (5xx), rate limits (429), or local network/browser problems (1020, some 4xx). A 1020 block is about your IP/network, not a ban. If you genuinely can't access your account after the standard fixes, contact OpenAI support.
Use a multi-model workspace like MultipleChat: when ChatGPT throws a 5xx, "at capacity" or other blocking error, you can switch to Claude, Gemini or another model and keep going for free instead of waiting for OpenAI to recover. It also lets you compare answers across models.
Error messages, codes and fixes change over time. Always confirm current behaviour on OpenAI's official pages — status.openai.com for live incidents and help.openai.com for troubleshooting articles. This guide is independent and not operated by OpenAI.