Over the past two days,I have repeatefly encountered response-stream falures in ChatGPT Projects while using GPT-5.6 Sol High.
The main problem is “Error in message stream.” In several cases, a response made substantial progress and appeared to be reasoning normally, but was cut off partway through. I have also encountered related failures such as “Something went wrong,” “Thinking Failed,” connection interruptions, or message-delivery timeouts.
Starting a branch did not resolve the problem. In the affected branch, three attempts at a complex text request produced very different results:
The first attempt produced a complete response, but it conflicted with previously established conversation context.
The second attempt replied almost immediately with only an acknowledgement that the task would be completed, but did not perform the requested work.
After regenerating, the third attempt visibly reasoned for approximately two minutes and produced a much more context-aware response, but the answer was cut off halfway with “Error in message stream.”
Other newly created chats inside the same Project have also shown errors. However, Work remains usable on the same account, and comparable tasks can still complete there.
The affected conversations are not exceptionally long by message count: one had approximately 16 messages when the stream error occurred, while the branch had approximately 35. This does not appear to be explained solely by a conversation naturally reaching an extreme length.
Expected behavior: Project chats and branches should complete responses consistently and preserve the context available before the branch point.
Actual behavior: reasoning depth, context application, task execution, and response streaming vary significantly between retries, with some responses stopping before execution and others failing after substantial progress.
Has anyone else reprodeced thie specifically in Projects or branched chats while Work continues to function normally?
I noticed this today too! I haven’t tried it in any other project folders, but it’s really annoying that I can’t use my custom project instructions and context for what I’m working on.
Also, my phone and computer are showing the same error. I updated the app on my phone, restarted the app multiple times, logged out and back in, and turned off my wifi (all suggestions from Google regarding this error).
Thanks — your cross-device result is useful. It makes a purely browser-specific explanation less likely.
For me, this has been ongoing for roughly the past two days. I have only tested the web version on desktop, so I cannot cannot confirm whether the same issue occurs in the mobile app.
The problem is concentrated around Project chats and branched conversations. Longer generation attempts can begin normally, then stop mid-response with “Error in message stream.” Some retries instead return an immediate acknowledgement or plan without actually completing the requested work.
Things I have already tried:
refreshing and reopening the conversation;
clearing browser cache;
starting new chats;
branching from an earlier point;
moving work to a new Project;
comparing the same kind of task in Work mode.
The behavior is inconsistent: some attempts produce a complete but context-conflicting response, some give a near-instant “I’ll handle it” response without execution, and some begin a longer, more context-aware response before failing in the message stream.
I’m seeing a very similar issue today, but in my case it appears isolated to a single Project.
Chats outside that Project work normally. Inside the affected Project, however, every existing chat and every newly created chat fails with either “Error in message stream” or “Something went wrong,” even for a simple prompt such as “hi”.
I reproduced the same behavior in an incognito browser session, so it does not appear to be browser-session specific.
I can also identify exactly when it started. The Project was working normally until I pasted an image directly into one of its chats and submitted the message. The image upload did not complete, and from that point onward the entire Project became unusable.
I deleted the conversation in which that happened, but that did not resolve it. Newly created chats inside the same Project still fail immediately.
I can’t say that the failed image upload caused the issue, but the timing was exact. This makes me wonder whether a failed attachment transaction or Project-level context/indexing state can leave a Project in a bad backend state.
I’m avoiding deleting or recreating the Project because it contains important Project Sources and accumulated context.
Getting same issue on project chats. Using Chatgpt on mobile app. New chats work fine. I need to use my project due to the sources. Anyone looking into this from @OpenAI_Support ?
I have the same issue on one project, fails on both browser and mobile app. It started with only new chats/branches having the issue but now its every chat in my project failing to generate a response.
I think I had the same issues while using ChatGPT.
My account is Plus, and I was using 5.6 Sol High for doing some pdf editing and long-context conversation. And it started to fail generating response regardless of new or old chats from yesterday.
I had the same problems on both Chrome and Microsoft Edge. I deleted all the caches, created a new profile in Chrome, diabled all the extensions, yet it continued to fail on generating response and showed me “Error in message stream.”
Update:
I tried to generate response on the chats, and the same “Error in message stream” happened.
I guess the issue appears tied to the regular Project-chat association or initialization state. I am still seeing the issue in regular Project chats and branches.
One additional control may be useful: I moved the same affected branch into Work mode because I did not want to abandon the context already accumulated in that conversation. After moving it to Work, that existing branch became usable again and I was able to continue working in it normally.
Update:
The remove-from-Project → send a simple message → move-back-to-Project sequence appears helpful for the explicit Project “Error in message stream” problem. It restored basic operation for the affected Project chat in my test.
However, after applying the workaround to the previously affected branch, I found that it did not resolve the branch-response problem described above. I have not yet done further testing, but this distinction matters for work that depends on accurate long-running conversation context: receiving a response is not enough if the response can still ignore established context or fail to execute the requested task.
Did not work for me. I remove the chat from the project and outside of it the “hi” message responds but when I bring it back to the project it gets the same “Error in message stream”
Update : I tried to add chats that had the ‘Error in message stream’ in new project, then it started to work fine for some time. But it got the same error when contexts were acculmated. I tried to remove that chat from the project and re-add it on the same project, but it was still same.
Weird thing is that this error happened inconsistently to me after I tried to add errored chat in a new project. But I think it is likely to happen when there is a long context that model had to calculate or when the model have to interact and run codes several times or the model was producing long response. But I can’t sure because sometimes chat stops even when it was just starting to use thinking and reasoning to generate response.
I created a branch from an earlier point in a conversation that had already been affected by the error, then moved that branch into a newly created Project. During the first few exchanges, the responses were not very good, but the chat could still continue and generally follow my instructions.
Starting with approximately the fifth exchange, however, the new branch became affected as well. The model appeared to begin thinking through the request, then produced only a few lines of an incomplete answer before the response ended with “Error in message stream.”
Ok! So, I’ve been dealing with this for the last 24 hours like many of you and have tried everything, but refuse to give up on my project because it is months of combined work and context.
I refreshed, cleared cache (via dev tools), new browser, new device, incognito, new wifi, logged in and out. Nothing works.
Just tried the removing chat from the affected project, still didn’t work. So I took that same chat, unpinned it, then I RENAMED THE CHAT. Then I gave the browser a refresh and it is back and working! I tried this in multiple chats that were all bricked within the project, they ALL came back to life, but ONLY after the refresh. I tried to talk to the chat after the rename and it errored out, but then I did a quick refresh and its back.
Free plan account. Accessing from web version and android mobile app.
All projects are impacted. If I remove a conversation from its project, it works; move it back into project, error in message stream or something’s wrong error messages return.
What model I use doesn’t matter.
I use projects for virtually everything, including health matters. Including ChatGPT for Health. The context built up in each project is essential.
Put simply, with projects, ChatGPT is useful enough that even though I’m permanently disabled and basically legally required to be poor, I’ve considered upgrading to a paid plan.
Without projects functioning, ChatGPT is useless. Completely and totally useless for anything.
Please see the fix that worked for me. I posted right before you.
I took all of my affected chats from the project, put them in a new project folder after I renamed them. Re-uploaded all the source material to the new project. Everything is working normally in a fresh project folder. Original chats and all.
I think this worked somehow. One of my chat that I renamed and moved to a new project acutally succeeded and produced a long response with reasoning which occured ‘Error in message stream’ before.
I’m trying to test the chat further to see if it is stable now.
ZachD85, thanks for the reply. I’ve read what you’ve posted.
Before I do as you suggest, though, I do have a concern.
I’ll acknowledge that I’m not a professional in this area, and I may be misunderstanding how Projects actually work.
The Projects I’m concerned about have been ongoing since January 2025 or so, maybe a bit later. There is a lot of Project context and memory built up over that time. I think of that, perhaps imperfectly, as an LLM equivalent of implicit context: not just the individual chats or uploaded files, but the accumulated contextual understanding that makes the Project useful as an ongoing workspace.
As I understand your method, creating a new Project and re-uploading the source material would preserve the chats and sources, but would leave behind whatever Project-level memory/context has accumulated in the original Project.
If that’s actually how it works, I’m concerned that the workaround could be worse than the problem it’s trying to solve. One of the Projects in particular concerns an extensive, unusual, and complex medical situation, and losing years of accumulated contextual continuity would be a significant problem; without the continuity, things could be misinterpreted in subtle ways that might be undetectable until they add up into a dangerous failure.
If I’m misunderstanding how Project memory/context works, I’d genuinely appreciate being corrected. If anyone knows of documentation or other material that explains what is actually preserved—or not preserved—when chats are moved to a new Project, I’d be grateful for something I can read, especially if it’s aimed at someone with a layperson’s level of technical knowledge.