The web version of ChatGPT has become incredibly slow today; I often wait an hour or two only to find that the “chain of thought” process has advanced just one or two steps before eventually returning a “message timed out” error. I have tried switching browsers and opening new windows, but the issue persists.
While response times have been sluggish since September, it used to at least generate a few decent answers. Today, however, it is completely failing; no matter how many new windows I try, the chain of thought almost always stalls after just three or four steps.
I am wondering whether you are working on a long-running task.
OpenAI now has documentation specifically for long-running work. For work that may require many steps, OpenAI recommends defining a clear outcome, constraints, and completion criteria.
In the ChatGPT desktop app, Codex CLI, and Codex IDE extension, /goal can be used for this type of work. There is also a more detailed Using Goals in Codex example. I have personally used /goal to keep a single objective going over several days.
For ChatGPT on the web, the documentation instead recommends ChatGPT Work for hosted long-running work.
I am also finding that GPT-6 Astra can remain focused on a task for a few hours without additional input from me. OpenAI’s model guidance specifically notes that Astra is better than earlier models at staying coherent during long tasks.
I have not yet given Astra one of my longer multi-day tasks. The distinction is also becoming less clear because ChatGPT Work and Codex now support subagent workflows, allowing independent parts of a larger task to be delegated and run in parallel.
That said, I would not assume that long-running work is the cause of the slowdown you are seeing. Your report of one- or two-hour waits followed by a timeout could be a separate performance or service issue.
I can barely use ChatGPT since early this morning. It stays in “thinking” mode for a very long time, and then comes back with Connection interrupted. Waiting for the complete answer.
It’s basically unusable. The status page says everything is fine but this is not true.
4o? 3.5-turbo? what is this bs site lel
Long conversations are partly broken since morning, suggestin to use work instead of gpt s like “use uzi instead of m4”. Work s useless in plus plan.
I may be seeing a related issue today, although in my case it goes beyond general slowness.
GPT-5.6 Sol High long-running tasks become progressively unreliable and eventually lose the response stream. In my latest clean reproduction today (Sep 17), the task ran for several minutes, then ChatGPT attempted to resume the stream, and /backend-api/f/conversation/resume eventually returned:
HTTP 404
{“detail”:“Resume stream unavailable”}
The client then fell back to polling and ultimately surfaced a network error without delivering the final answer.
I have reproduced this on ChatGPT Web, the Windows app, and the Android app over a separate cellular connection, so it does not appear to be limited to one browser or one network.
I already posted the full details and diagnostics in this related thread:
I’m mentioning it here because the timing overlaps with the broader web slowness reports today, and these issues may be related.
I’m glad I’m not the only person having the exact same issue with ChatGPT since yesterday.
I’ve tried on every device and every browser you can think of, and “stopped thinking” for any long-running tasks.
It doesn’t even last five minutes before displaying the message, rendering any chat unusable if you need it to reason for more than a couple of minutes.
Recently for the past few days, even when I make a new chat, the responses have been taking over 30 minutes without even responding. Before it would be so much faster which wasn’t that long ago. It’s definitely having the issues rn.
It’s abysmally slow right now, taking several minutes per thinking step almost 90% of the time. It even takes over a minute sometimes on simpler non-programming stuff, and it’s kind of annoying me. 5.6 wasn’t this horrifically slow on launch.