ChatGPT has 1.2 billion weekly users, so when it breaks, it breaks for a lot of people at once. But most “ChatGPT not working” moments are not full outages. They come from a VPN, a browser extension, a wrong device clock, an overloaded model or a stuck chat. This guide starts with a 60-second check to tell an outage from a local problem, then lists the exact error messages ChatGPT shows in 2026 with the cause and fix for each. It covers the web app, the desktop and mobile apps, and Codex.
- Check status.openai.com. If there is an incident, wait; local fixes will not help.
- Refresh, or start a new chat. Many errors are tied to one broken conversation.
- Turn off VPN, proxy and ad or script blockers. They cause most network and “unusual activity” errors.
- Try a private window, another browser, or the app instead of the website.
- Clear ChatGPT’s cookies and cache, then log in again.
- On phones, set the date and time to automatic and update the app.
- If one model says it is at capacity, switch model or start a fresh chat.
Is ChatGPT Down? The 60-Second Check
- Open status.openai.com. Look at ChatGPT components (Conversations, Login, Voice, File uploads, Codex). A yellow or red bar means it is not you.
- Try ChatGPT on another device or network, for example your phone on mobile data. If it works there, the problem is your computer or network.
- Check whether only one feature fails (images, files, voice). Partial incidents are common: plain text chat may work while tools are down.
If the status page shows an incident, the best fix is to wait and then start a new chat once it is resolved. Retrying hard during an outage can trigger rate-limit messages.
“Something Went Wrong” Errors
“Something went wrong” / “Hmm…something seems to have gone wrong.”
Cause: a temporary server problem or a local setup issue. The second wording is not documented by OpenAI but shows up during outages.
Fix: refresh; start a new chat; check the status page; clear cookies and cache for chatgpt.com; try a private window; disable extensions, VPNs and secure-DNS settings one by one.
“Something went wrong. Please make sure your device’s date and time are set properly” (mobile)
Cause: the app could not create a security token, usually because the phone’s clock is off.
Fix: set date and time to automatic, update the ChatGPT app and your OS, then reopen the app.
“There was a problem preparing your chat”
Cause: an extension or script blocker stopped the page from loading.
Fix: allow chatgpt.com in your blocker, or disable script-blocking extensions.
Network and WebSocket Errors
“A network error occurred. Please check your connection and try again.”
“An error occurred while connecting to the websocket”
Cause: your device cannot keep a steady connection to OpenAI, very often because a VPN, proxy, firewall or corporate filter blocks WebSocket traffic. Long answers are more likely to fail.
Fix:
- Turn off VPNs and proxies, or switch the VPN to a different server.
- Try another browser or network (mobile hotspot is a quick test).
- On a work or school network, ask IT to allow WebSocket connections to chatgpt.com and openai.com.
- Ask for shorter answers or split a big task into parts.
Response and Streaming Errors
“There was an error generating a response”
Fix: click Regenerate. If it keeps failing, restart the browser or app, disable VPNs and proxies, or try another device.
“Error in message stream”
Cause: the answer’s stream broke halfway. Fix: regenerate, and if the chat is very long, start a new one and paste a short summary of the context.
Stuck on “Thinking” forever
Fix: stop the response, close duplicate tabs of ChatGPT, and retry. Reasoning models can take minutes on hard prompts, but if nothing appears after several minutes, stop and switch to a faster model.
Too Many Requests and Capacity Errors
“Too many requests. You’re making requests too quickly…”
Cause: rate limiting, often during incidents or when scripts, extensions or many tabs send requests at once. Plain text chat is reported to have been unlimited on paid plans since August 6, 2026, so this message usually signals a backend issue rather than your plan cap.
Fix: wait a few minutes, close extra tabs, stop running agent or Work tasks, and check the status page. Contact support if it lasts for days.
“Too many concurrent requests”
Fix: close duplicate tabs, stop background tasks, and wait.
“Selected model is at capacity. Please try a different model”
Cause: the model you picked is overloaded. Fix: start a fresh chat with the same model (OpenAI support’s suggested fix), or switch to another model for now.
Usage limit messages
If you see a message that you have reached your limit for a model or feature, that is your plan’s cap, not an error. It shows when the limit resets. Compare plan limits in my ChatGPT Free vs Go vs Plus vs Pro guide. Developers seeing 429 errors on the API should check rate limits and billing; my OpenAI API pricing guide covers tiers.
Account, Login and Unusual Activity
“Unusual activity has been detected from your device”
Cause: traffic that looks automated, very often from a shared VPN or proxy IP address. Fix: turn off the VPN or proxy, restart the browser or use a private window, and change your password. If you do not use any automation and it continues, contact OpenAI support.
Suspicious activity alert (“more simultaneous logins than usual”)
Fix: turn on two-factor authentication, then use “log out of all devices” in Settings and sign back in. Do not share your account.
Missing Chats and History Errors
- “Unable to load history”: the sidebar failed to load; your chats are not deleted. Click Retry, use Search, or open the desktop or mobile app.
- “You don’t have access to this conversation”: the chat belongs to another account or workspace, or the shared link was deleted. Log in to the right account and check Archived chats.
- “Conversation not found”: deleted, in another account, or it was a Temporary Chat (those are not saved).
Files and Tools Errors
- “Some tools are temporarily unavailable”: the tool layer (file creation, uploads, code, images) has an outage while text still works. Check the status page and start a new chat once it recovers.
- “Download failed” / “File not found”: the generated file expired or was over 512 MB. Ask ChatGPT to regenerate it.
- Image generation fails: often a temporary tool outage, or the prompt hit a content rule. Rephrase, or wait if the status page shows image issues. More on the latest image model in my ChatGPT Images 2.6 guide.
Codex Errors
- “Failed to read configuration layers”: Codex could not read
~/.codex/config.toml, usually because of a TOML typo, duplicate keys or a damaged file. Back it up, fix the line the error points to, or restore a backup. - “Permission denied (os error 13)”: you ran Codex with sudo, so
~/.codexbecame owned by root. Give the folder back to your user (for examplesudo chown -R $USER ~/.codexon macOS or Linux).
More on Codex limits and setup in my OpenAI Codex guide.
All Errors at a Glance
| Error | Most likely cause | First fix |
|---|---|---|
| Something went wrong | Server blip or browser setup | New chat, then clear cache |
| Date and time error (mobile) | Wrong device clock | Set time to automatic |
| Network error / websocket | VPN, proxy, firewall | Disable VPN, try another network |
| Error generating a response | One failed answer | Regenerate |
| Error in message stream | Broken stream | Regenerate or new chat |
| Too many requests | Incident or too many tabs | Wait, close tabs |
| Model at capacity | Model overloaded | New chat or other model |
| Unusual activity | Shared VPN IP | Disable VPN, change password |
| Problem preparing your chat | Extension blocking scripts | Disable blocker |
| Unable to load history | Sidebar failed | Retry or use the app |
| Tools temporarily unavailable | Tool outage | Wait, check status |
| Download failed | File expired or too big | Regenerate the file |
How to Avoid Most ChatGPT Errors
- Keep one ChatGPT tab open, not five.
- Start new chats for new topics; very long chats fail more often.
- Add chatgpt.com to your VPN’s split-tunnel exclusions.
- Keep the desktop and mobile apps updated.
- Have a backup assistant ready for outages. Claude, Gemini or a local model all work; see my ChatGPT vs Claude vs Gemini vs Grok comparison or run one offline with Ollama.
Using Claude and seeing errors there instead? See my Claude error fix guide.
Frequently Asked Questions
Why does ChatGPT keep saying “Something went wrong”?
Usually a short server issue or a browser problem. Start a new chat, check status.openai.com, then clear cookies and disable extensions and VPNs.
How do I know if ChatGPT is down right now?
Check status.openai.com for incidents and test on another device or network. If it fails everywhere and the status page shows an incident, it is down.
Why does ChatGPT work on my phone but not my PC?
Something on the PC is interfering: a VPN, proxy, firewall, extension or stale cookies. Try a private window first.
Does a VPN break ChatGPT?
It can. Shared VPN IPs trigger “unusual activity” checks, and some VPNs break WebSocket connections, causing network errors.
Are my chats deleted if ChatGPT says “Unable to load history”?
No. That message means the sidebar failed to load, not that your chats are gone. Click Retry, use Search, or open the desktop or mobile app. If one chat is missing, check Archived chats and make sure you are in the right account. Temporary Chats are never saved.
Why is ChatGPT stuck on “Thinking”?
Reasoning models can take a few minutes on hard prompts. If nothing appears after several minutes, stop the response, close duplicate ChatGPT tabs, and retry or switch to a faster model.
Why can’t I download a file ChatGPT created?
Generated files expire after a while, and very large files can fail. Ask ChatGPT to regenerate the file in the same chat, then download it straight away.
