I’m seeing what appears to be the same GPT-5.6 Sol High long-running execution / stream-resume problem.
I want to keep this separate from the “Suspicious activity” issue itself: I’m specifically reporting the long-task execution / streaming failure.
I’m a ChatGPT Plus subscriber.
My pattern is very consistent:
- GPT-5.6 Sol High starts the task normally.
- It successfully reads uploaded files and performs analysis for several minutes.
- The task is clearly progressing normally.
- Then the response stream stops.
- ChatGPT attempts to resume the generation.
- The resume eventually fails and no final assistant response is delivered.
I have reproduced this across:
- ChatGPT Web
- the official ChatGPT Windows app
- the official ChatGPT Android app
- my home internet connection
- a completely separate cellular/mobile-data connection
- multiple conversations/projects
- and a brand-new conversation
Simple prompts work normally.
I also tested a similar workload with GPT-5.6 Sol Medium, and that was able to complete, while the longer High-reasoning runs repeatedly fail.
The most recent clean reproduction was on September 17, 2026.
In a brand-new GPT-5.6 Sol High conversation, the main run started at approximately 12:35 CEST.
After several minutes, ChatGPT attempted to resume the stream.
One resume connection remained open for about 93 seconds and received keepalive pings, but no final completion arrived.
A subsequent:
/backend-api/f/conversation/resume
returned:
HTTP 404
{“detail”:“Resume stream unavailable”}
at approximately 12:42:59 CEST.
The client then fell back to polling and eventually surfaced:
“A network error occurred. Please check your connection and try again.”
No final assistant message was delivered.
An earlier reproduction also showed:
net::ERR_HTTP2_PROTOCOL_ERROR
followed by:
/backend-api/f/conversation/resume → 404
{“detail”:“Resume stream unavailable”}
The Windows desktop app has also directly shown:
“Resume stream unavailable”
And on the official Android app, using cellular data rather than my home network, the High-reasoning generation stopped without ever delivering the final response.
So at least in my case this does not appear to be limited to a specific browser or a single internet connection.
I have already collected sanitized HAR captures, browser console logs, timestamps, x-oai-request-ids, cf-ray values, and turn trace IDs and provided the relevant diagnostics to OpenAI Support.
I am not posting the HAR/log files publicly because they contain session/client diagnostic information, but I can provide them directly to OpenAI Support/engineering if needed.
For reference, my setup is:
Plan:
ChatGPT Plus
Model:
GPT-5.6 Sol
Reasoning:
High
Windows ChatGPT app:
26.911.61220
Android ChatGPT app:
1.2026.251 (11)
Phone:
POCO X6 Pro 5G
Android 16
Browser used for the web HAR reproduction:
Brave 1.95.102
OS:
Windows 11
The important part for me is that the same underlying failure has now reproduced across the web client, the official Windows application, and the official Android application over a separate cellular network.
I’m adding my case here because it looks very similar to the long-running GPT-5.6 High/Extra High failures other users in this thread have described.
If OpenAI staff/engineering are tracking this, I can provide the exact request IDs, trace ID, timestamps, HAR, and console logs through Support.
