Error messages

Error in message stream

Error in message stream

ChatGPT shows this under a reply that stopped partway, or in place of one, usually with a retry button:

Error in message stream

ChatGPT sends its answer to you in pieces as it's written (that's the "stream"). This line means the stream broke before the answer was done. It shows up on the web, in the desktop apps and on Android, and there's a steady run of forum threads about it: five new ones on OpenAI's community forum in September 2026 alone.

What OpenAI says it means

Not much, and they say so. An OpenAI_Support account answered one thread on September 8, 2026 with "This is a generic error message could be due to multiple reason", and asked for a HAR file (a recording of the browser's network traffic) plus the time in UTC. OpenAI's troubleshooting page for ChatGPT errors doesn't list this message at all as of September 25, 2026.

It has also been a plain outage more than once. In November 2025 developers testing ChatGPT apps built with the Apps SDK got it on nearly every request, and OpenAI Support posted that a fix went out on November 19 and was confirmed the next day.

When it sticks to one chat

The pattern that keeps coming up is a single conversation that fails every time while a new chat works fine. In a June 2026 thread a forum moderator asked "Is it a long thread?" and the person said yes, 30-plus requests; branching the chat failed too. OpenAI Support's reply in September: "We haven't verified a fix for this exact issue." They asked for the model, browser, time and the affected chat link, sent privately to Support.

The help page's advice for slow or stuck replies fits this case: "Start a new chat if the conversation is long or has many turns." Reports from September 2026 describe the same split with GPT-5.6 Sol on High reasoning, where short requests work and long reasoning runs die partway through.

When a whole Project fails

Since about September 6, 2026 a bigger thread has collected people whose every chat inside one ChatGPT Project fails (with this line or "Something went wrong"), even for "hi", on web and mobile, while chats outside the Project work. Two workarounds came out of it:

• Remove the chat from the Project and send from there. Some people had it work until they moved the chat back.
• Create a new Project, move the chats into it, and upload the source files again. Several people confirmed this one.

Neither held for everyone. Two people who'd had it working said by September 24 and 25 that the error was back, so treat both as stopgaps. One poster relayed that Support considered the problem isolated to the affected Project and still under review as of September 13. Keep the original Project and its files until it's settled.

What to try, in order

Retry once. Check status.openai.com, since an incident breaks every chat at once. If it's one chat, start a new one and paste in what it needs. If it's one Project, try the moves above. If it follows you across browsers, devices and networks, OpenAI's help page asks for a HAR file, console errors with timestamps, the model and the conversation URL when you contact Support.

What surprised us: the generic advice people got back from Support (refresh, turn off the VPN, try another network) doesn't match what the forum found, which is that the trouble travels with the conversation or the Project, on whatever device you open it.