ChatGPT Error in Message Stream: Causes and Fixes (2026)
"Error in message stream" means ChatGPT's answer stopped arriving partway: the stream of text from OpenAI's servers to your app broke before the reply finished. It is usually server-side, during an incident, on a very long reasoning run, or in one broken chat or Project. Click Retry first; if the same chat keeps failing, start a new chat; then check status.openai.com and turn off VPNs and secure DNS (OpenAI Status and Help Center, 2026).
OpenAI has no Help Center article for this exact message. Its own support team calls it "a generic error message" with several possible causes, and the only status incident named after it, on June 11, 2025, was fixed by rolling back a change. This guide maps each pattern people report to the fix that matches it, using OpenAI's status page, release notes and troubleshooting articles as of October 2026 plus OpenAI support replies in the Developer Community, and it says which is which. The last sections cover keeping long work safe when a stream breaks, including with AI Toolbox (formerly ChatGPT Toolbox), a Chrome extension with 40,000+ users, a 4.7/5 Chrome Web Store rating, and a free plan.
By Adi Leviim, Co-Founder at Infi Developments. 7+ years building AI productivity tools. Last reviewed October 2026.
What "Error in Message Stream" Means in ChatGPT
ChatGPT sends each answer as a stream, word by word, over a long-running connection; "Error in message stream" is what you see when that stream ends with an error instead of a finished reply. The partial answer usually stays on screen, cut off, with a Retry option underneath. The message tells you where it broke, not why, which is why OpenAI support describes it as "a generic error message could be due to multiple reason" (OpenAI Developer Community, September 8, 2026).
When you see it
Likely cause
What to try
In every chat, suddenly, along with other people
An OpenAI incident
Check status.openai.com and wait
In one long chat, every time, while new chats work
A long thread that fails to stream
Start a new chat with a summary
Minutes into a long Thinking or High-effort answer
A long reasoning run that was cut off
Lower the effort or split the task
In every chat inside one Project, even "hi"
A problem with that Project, sometimes after a failed upload
Try a new chat, then move chats out of the Project
Only on a work or school network
WebSocket or long connections blocked or cut
Try mobile data; ask IT to allow ws.chatgpt.com
In ChatGPT Work, after a long time on Thinking
A Work mode issue
Use regular chat, check status
OpenAI has acknowledged the error directly twice. On June 11, 2025 it reported "an increase in 'Error in message stream' errors in ChatGPT" and rolled back the change behind it within 25 minutes (OpenAI Status). Its February 27, 2026 release notes list "Fixed increased message streaming errors in GPT-5.2 Thinking" (ChatGPT release notes).
How to Fix "Error in Message Stream"
Work through these in order, quickest first. None of them deletes a chat.
Retry the answer. Use the Retry button under the failed reply, or the three dots under a Thinking or Pro response and choose to regenerate. Since March 2025 OpenAI's release notes say that after a message error "you can retry or just continue chatting in the same conversation." On Android, use Try again.
Start a new chat if the same chat fails again. Users with long threads (one had "30+ requests in the chat") found the old chat and even branches of it failed immediately, while a new chat worked. OpenAI support confirmed that pattern on September 10, 2026: "branching also fails and that a new chat works."
Check status.openai.com. If an incident is open, especially "elevated errors" in ChatGPT conversations, wait for it to resolve.
Make the request smaller. On a long reasoning request, choose a lower effort or split the job into steps.
Refresh hard and sign back in. OpenAI's sequence for a stuck answer includes a hard refresh with Cmd or Ctrl+Shift+R and signing out and back in (OpenAI Help Center).
Open up the connection. Turn off VPNs, proxies and secure DNS tools like Web Protect, try a private window with extensions off, or switch to mobile data.
Contact OpenAI Support if it continues across browsers, devices and networks. Record a HAR file while the error happens and note the time in UTC, the model and the chat link; OpenAI support asks for exactly these.
Clearing your cache rarely helps with this particular error: users in several threads cleared cache and cookies with no change, because the failure was on the server or in that specific chat. For blank screens and endless spinners instead, see the ChatGPT freezing guide.
Long Chats and Long Reasoning Runs
The two most common patterns in 2026 reports are a chat that has grown very long and a reasoning answer that runs for many minutes.
Long chats: once a thread fails this way, users report every new message in it fails too, even after branching. OpenAI's general advice for slow or failing chats is to "start a new chat if the conversation is long or has many turns."
Long reasoning: a Plus user on GPT-5.6 Sol with High reasoning reported on September 22, 2026 that "short and simple requests generally work," while complex ones ran for several minutes and then stopped with the error. Others in a long September thread described runs ending at about 25 to 26 minutes. OpenAI has not published a time limit, so treat this as reported, not documented.
Model-specific waves: earlier reports tied it to one model at a time, such as o4-mini-high in May 2025 and o3-pro on the Mac app in June 2025, and the February 2026 fix named GPT-5.2 Thinking. If only one model fails, switch models for that task.
Starting fresh does not have to mean starting over. Ask ChatGPT for a summary of the decisions so far before the thread breaks, or paste your own notes into the new chat. The ChatGPT context window guide explains how much a chat can hold on each plan.
"Error in Message Stream" in ChatGPT Projects
If every chat inside one Project fails, even a one-word prompt, while chats outside it work, the Project itself is the problem. A 38-post thread from September 2026 describes exactly this, reproducible in private windows, other browsers and the mobile app (OpenAI Developer Community). One user traced it to an image paste whose upload did not complete, after which "the entire Project became unusable."
On October 1, 2026, OpenAI support replied: "We fixed a related issue where an unusable Project source could stop a chat before its first reply. Can you try a new chat in the affected Project?" Some users still saw failures afterwards. What users tried, none of it official:
Start a new chat in the Project after the October 1 fix.
Remove a failing chat from the Project, send a message, then move it back.
Move the chats to a new Project and add the sources again.
Remove the Project's sources and instructions, then add them back one at a time to find the one that breaks it.
Before reorganizing a Project, export the chats you need from it. The ChatGPT Projects guide covers moving chats and sources.
Long chats break. Your work does not have to. AI Toolbox (formerly ChatGPT Toolbox) shows how full a ChatGPT chat is with its free Context Window Meter, and on Premium carries a summary of a long conversation into a fresh chat in one step. Export any chat as TXT free, or Markdown, JSON and PDF on Premium. 40,000+ users, 4.7/5 on the Chrome Web Store.
A network that cuts long-running connections can end a stream early, so the error is more common on corporate networks, VPNs and filtered DNS. OpenAI's network article says ChatGPT uses a WebSocket connection to wss://ws.chatgpt.com for conversation updates, and that "if your organization blocks WebSocket traffic by default, some features may stall, disconnect, or fail to stream updates correctly" (OpenAI Help Center).
Quick test: OpenAI suggests checking whether the error goes away on a phone hotspot instead of the company Wi-Fi.
For IT teams: allow the ChatGPT domains and WebSocket over TCP 443, and if sessions "connect but later stall or disconnect, check idle timeout and maximum frame or message-size policies for WebSocket traffic." TLS inspection must not close the long-running WebSocket connection.
At home: turn off VPNs, proxies and secure DNS tools while you test.
If the error appears on every network and device, the cause is not your connection. One September 2026 report put it plainly: "Not an extended chat, a large file, or an unstable internet connection."
What Happens to the Partial Answer
The text that arrived before the break usually stays visible, cut off, and the full answer is not saved. Retrying generates a new answer rather than resuming the old one.
Retry is built in: the failed reply shows a Retry button, and you can keep chatting in the same conversation.
Turns can go unsaved during incidents: during an August 4, 2026 incident OpenAI said "some conversation turns may not be saved" (OpenAI Status).
Long Thinking answers can vanish: users have reported a refresh leaving only "Thought for 13m" and an empty reply.
Copy what you can. If a partial answer contains something you need, copy it before you retry or refresh.
OpenAI does not document whether a partial answer is always kept in your history, so the safe habit is to keep your own copy of long, valuable answers as soon as they arrive.
On the Desktop and Mobile Apps
The error is not limited to the browser. Users report it in the Windows desktop app (in French as "Erreur dans le flux de messages," with a Retry button), the Mac app, and the Android app, and in Projects on both phone and computer. OpenAI's Help Center does not list which apps show it, and we found no report naming the iPhone app specifically. In the apps the same order applies: retry, new chat, status page, then a different network. On Android, update the app from Google Play; on desktop, reload the app.
Developers building ChatGPT apps or custom connectors saw a separate wave in November 2025, which OpenAI fixed on its side ("We can now confirm that this issue has been resolved"). If the error appears only when your own app or connector runs, refresh ChatGPT and test the tool on its own.
Native ChatGPT vs AI Toolbox for Long, Fragile Chats
AI Toolbox cannot stop OpenAI's servers from dropping a stream; it helps you notice a chat getting long and keep a copy of what matters.
Task
ChatGPT (built in)
AI Toolbox Free
AI Toolbox Premium
See how full a long chat is
Not shown
Context Window Meter
Context Window Meter
Move to a fresh chat without losing context
Ask for a summary and paste it
Not included
Summary handoff to a new chat in one step
Keep a copy of a long answer
Copy by hand; full account export by email
Export the chat as TXT
TXT, Markdown, JSON, PDF, in bulk
Find the chat again later
Search titles and messages
Full-text search, first 5 results
All results plus Advanced Search filters
Price
Included
Free
$9.99/month or $99 lifetime
The meter estimates the conversation's text and does not count tokens from uploaded files, so treat it as a floor in file-heavy chats. Exports are made from the conversation ChatGPT has saved, so they include what ChatGPT kept, not text that never arrived. The ChatGPT export guide compares every export method.
Limitations
OpenAI documents this error only through one status incident, one release note and support replies, so most causes above come from user reports with dates rather than official explanations. Some cases have no user-side fix at all: a server incident, a regression in one model, or a broken Project waiting on OpenAI. Native ChatGPT handles the simple case well, because Retry fixes a one-off break in a single click. When the error persists everywhere, only OpenAI Support can trace it, using a HAR file and the time it happened.
Frequently Asked Questions
What does "Error in message stream" mean in ChatGPT?
It means the answer ChatGPT was streaming to you stopped with an error before it finished. The cause is usually server-side, such as an incident, a very long reasoning run or a problem with one chat or Project, and sometimes a network that cuts long connections. OpenAI support calls it a generic error with multiple possible causes.
How do I fix "Error in message stream" in ChatGPT?
Click Retry or Regenerate first. If the same chat keeps failing, start a new chat with a short summary of the old one. Check status.openai.com, lower the reasoning effort or split long tasks, and turn off VPNs, proxies and secure DNS or try mobile data. If it persists everywhere, contact OpenAI Support with a HAR file.
Why does ChatGPT keep saying error in message stream in one chat?
Long threads can get into a state where every new message fails, and users report branches of the chat fail too while new chats work. OpenAI support confirmed that pattern in September 2026. Start a new chat and carry over a summary, and export the old chat if you need its contents.
Why do all my ChatGPT Project chats show error in message stream?
When every chat in one Project fails and chats outside it work, the Project is the problem, sometimes after a failed upload. OpenAI said on October 1, 2026 that it fixed a related issue with unusable Project sources. Try a new chat in the Project, then move chats or sources to a new Project.
Is "Error in message stream" my internet connection?
Sometimes. Networks that block or time out WebSocket connections can cut a stream, so test on mobile data and turn off VPNs and secure DNS. If it fails on every network and device, or for many users at once, the cause is on OpenAI's side.
Do I lose the answer when I get error in message stream?
The partial text usually stays visible, but the complete answer is not saved, and retrying writes a new answer. During some incidents OpenAI says conversation turns may not be saved at all. Copy anything important before retrying, and export long chats you rely on.
Does error in message stream happen on the ChatGPT app?
Yes. Users report it on the Windows and Mac desktop apps and the Android app as well as the web. The fixes are the same: retry, start a new chat, check the status page, and try a different network. Updating or reloading the app can also help.
Bottom Line
"Error in message stream" is a broken delivery, not a broken account. Retry once, then move to a new chat if the same thread keeps failing, check OpenAI's status page, and shrink very long reasoning requests. For Projects, try a new chat after OpenAI's October fix before rebuilding. Keep copies of long answers so a dropped stream never costs you the work.
A Full Stack Developer with 7+ years of experience building AI productivity tools. Leads product development and frontend architecture for AI Toolbox, the Chrome extension suite (ChatGPT, Gemini, Grok, and Claude modules) that helps users search, organize, and export their AI conversations.
Liked this guide? Get the next one first.
Join 40,000+ readers getting practical prompts, new features, and subscriber-only deals. No spam, unsubscribe in one click.
New features and product updates before anyone else
Time-saving prompts and workflows for ChatGPT, Gemini, Claude, and Grok
Adi Leviim