ChatGPT: Error in Message Stream
ChatGPT sends every answer as a stream of text over a long-running connection. 'Error in message stream' appears when that stream ends with an error before the reply finishes, so you usually see a partial answer cut off with a Retry option. OpenAI support has described it as a generic error with several possible causes, and the message itself tells you where it broke, not why. The patterns users report fall into a handful of groups: an OpenAI incident (OpenAI's status page has logged incidents named after this exact error, and release notes have listed fixes for increased streaming errors in specific models); one very long chat that now fails on every message while new chats work; long Thinking or high-effort reasoning answers that run for many minutes and then stop; a single Project where every chat fails, even a one-word prompt; and networks that block or time out WebSocket connections. Retrying fixes one-off breaks, but repeat failures need the fix that matches the pattern.
Affiliate link, we may earn a commission at no extra cost to you.
Symptoms You Might See
An answer starts, then stops partway with 'Error in message stream'
Every new message in one long chat fails, but new chats work
Long Thinking or high reasoning answers stop after several minutes
Every chat inside one Project fails, even 'hi'
It only happens on a work or school network, or with a VPN on
Common Causes
An OpenAI incident or model regression
OpenAI has acknowledged spikes of this exact error on its status page and has shipped fixes for increased message streaming errors in specific models. When it hits many people at once, it is not your setup.
A very long chat
Users report that once a long thread fails this way, every new message and even branches of it fail, while new chats work. OpenAI support confirmed that pattern in September 2026.
Long reasoning runs
Complex requests on high reasoning settings can run for many minutes and then stop, while short requests in the same chat work. OpenAI has not published a time limit, so treat this as a reported pattern.
A broken Project
If every chat inside one Project fails, the Project itself is the problem, sometimes after an upload that did not complete. OpenAI said on October 1, 2026 that it fixed a related issue with unusable Project sources.
Networks that cut long connections
ChatGPT uses a WebSocket connection for conversation updates. OpenAI's network guidance notes that blocking or timing out WebSocket traffic can make streaming stall or fail, which is common on corporate networks and with some VPNs and filtered DNS.
Step-by-Step Fixes
Retry the answer
When to try: First, for a one-off failure
Use Retry under the failed reply, or regenerate from the menu under a Thinking response. OpenAI's release notes say you can retry or keep chatting in the same conversation after a message error. Copy any partial text you need first, because retrying writes a new answer.
Start a new chat with a summary
When to try: When one chat fails repeatedly
If the same chat keeps failing, start a new chat and paste a short summary of decisions so far (or ask for one before the thread breaks). This is the fix that works for long threads in user reports and OpenAI support replies.
Check status.openai.com
When to try: When it starts suddenly in every chat
Look for elevated errors in ChatGPT conversations. If there is an incident, wait for it to resolve instead of retrying long requests.
Shrink long reasoning requests
When to try: When long Thinking answers stop after minutes
Choose a lower reasoning effort, or split the job into smaller steps, for example outline first, then each section. If only one model fails, switch models for that task.
Fix a failing Project
When to try: When every chat in one Project fails
Try a new chat in the Project first. If chats still fail, move the chats you need to a new Project and add the sources back one at a time to find the one that breaks it. Export important chats before reorganising.
Open up the connection
When to try: When it only fails on one network
Turn off VPNs, proxies and secure DNS tools, try a private window with extensions off, or switch to mobile data. On company networks, ask IT to allow ChatGPT domains and WebSocket traffic on port 443 without short idle timeouts.
Contact OpenAI with evidence
When to try: When nothing above helps
If it continues across browsers, devices and networks, contact OpenAI support from help.openai.com with the time in UTC, the model, the chat link and, if asked, a HAR file recorded while the error happens.
How to Prevent This
Start new chats for new tasks instead of growing one huge thread
Keep a copy of long, valuable answers as soon as they arrive
Use lower reasoning effort for routine requests
Avoid VPNs and filtering DNS while working in ChatGPT
When to Contact Support
Contact OpenAI support when the error happens in new chats on several networks and devices with no incident on the status page, or when a whole Project stays unusable after OpenAI's fixes. Include the UTC time, model and chat link.
Frequently Asked Questions
What does 'Error in message stream' mean?▼
The answer ChatGPT was streaming to you stopped with an error before it finished. It is usually server-side, a very long chat or reasoning run, a broken Project, or a network that cuts long connections.
Why does one chat keep saying error in message stream?▼
Long threads can get into a state where every new message fails, and users report branches fail too while new chats work. Start a new chat with a summary of the old one.
Do I lose the answer?▼
The partial text usually stays on screen, but the complete answer is not saved and retrying writes a new one. Copy anything important before retrying.
Is it my internet?▼
Sometimes. Networks that block or time out WebSocket connections can cut the stream. If it fails on every network and device, or for many users at once, the cause is on OpenAI's side.
Does clearing the cache fix it?▼
Rarely. Users who cleared cache and cookies usually saw no change, because the failure was on the server or in that specific chat.
Sources and verification
Product behavior and limits can change. These primary sources were used to verify this guide.
Other Common AI Errors
ChatGPT Not Working
Service Issues
🚦Rate Limit Exceeded
Usage Limits
📏Context Length Exceeded
Input Limits
🔥Model Overloaded
Service Issues