You Asked It To Think. It Took 3 Seconds.
Since Saturday, paying ChatGPT users on the web say the High reasoning setting does nothing, short conversations are locking with a "Chat memory full" warning, and files will not download.
A wooden hourglass, shown for illustration. Photo: S Sepp, CC BY SA 3.0, via Wikimedia Commons.
OpenAI's status page has been quiet since Friday night. The OpenAI Community forum has not. Starting late Saturday Pacific time, paying ChatGPT users began posting the same complaint: they choose the High reasoning setting, and ChatGPT answers almost instantly anyway, with the kind of shallow reply that setting is supposed to prevent.
"Worked for 3 seconds"
At 12:09 AM Pacific on Sunday, October 11, a Plus subscriber posting as qiholite20 opened a thread on the forum. The setup, in the user's words: ChatGPT on the web in Chrome and Edge, "GPT-6 Sol / GPT-5.6 Sol + High reasoning effort." The symptom: "Even when High is explicitly selected, the model starts replying almost immediately with shallow answers. No visible extended thinking process." The user says it started around October 10 and survived a new chat, the retry button, different browsers, the desktop and mobile apps, and a cleared cache. According to the post, the request payload still showed the model and the high thinking effort that had been asked for.
Seven minutes later another user, jotpenAI, replied: "Even when I attach a lengthy report and explicitly tell GPT to read it, it sometimes replies after just “Worked for 3 seconds.”" The same user added: "I’m utterly exhausted from having to babysit it all the time." The original poster answered with "Same experience" and called the problem "exhausting." A third user wrote simply: "I’m having the same issue."
Later Sunday morning a user posting as dlocke5300 added a detail that points at the clients rather than the account. On the same account, the post says, the Android app still produces the longer reasoning it should, while ChatGPT on the web and the Windows desktop app "often return responses almost immediately, with noticeably less depth and reasoning quality." The user wrote that "Changing the reasoning level does not seem to have any meaningful effect," even after new conversations, different browsers, a cleared cache and reinstalling the desktop app.
Another user, working in ChatGPT Work in the middle of a software build, described a different version of the same frustration: a two minute delay before anything happens when it does think longer, then answers that "come out generic or not-attached to what we are working on," and a model that "seems to forget source files exist."
Short chats now say "memory full"
At 7:31 AM Pacific Sunday, a user posting as hulur084 started a thread titled "ChatGPT’s new “Chat memory full” is genuinely so annoying." The user says recent chats are being locked "even when they’re short, while older chats still work fine," and that "The send button just gets greyed out and I can’t do anything." The post asks why users are being forced to split conversations across multiple chats, says "the context transfer doesn’t even work properly half the time," and adds: "This wasn’t happening a few days ago."
Files that will not download
Earlier Sunday, at 5:45 AM Pacific, a user posting as Radu1913 reported being unable to download files, both ones ChatGPT itself produced and ones the user had uploaded, from the "View files in chat" section. The post ends with: "Please solve this ASAP."
And at 2:33 AM Pacific, a user posting as Damokkles filed a GPT-6 reliability report saying ChatGPT "repeatedly asks for information already provided in previous conversations, sometimes claims to remember details without verification, and repeatedly promises improvements without demonstrating consistent changes."
What OpenAI has said
Nothing, as of Sunday morning. The most recent incidents on status.openai.com are the three from Friday, October 9, which we covered yesterday. None of the threads above had an OpenAI reply when we read them. We can't confirm from the outside whether High requests are being routed to a lighter model or whether this is a client bug, but the pattern users describe is consistent: the setting is selected, the payload says high, and the answer arrives in seconds.
What you can do
If you rely on High reasoning, one user in the thread says the Android app still behaves normally, so it may be worth comparing the same prompt there before trusting a three second answer on the web. If a chat locks with "Chat memory full," copy anything important out before starting a new one, since users say the context transfer is unreliable. And if this is happening to you, read what others are reporting.
Sources: OpenAI Community, High reasoning selected but replies instantly with no thinking; OpenAI Community, ChatGPT’s new “Chat memory full” is genuinely so annoying; OpenAI Community, Can’t download files at all; OpenAI Community, GPT-6 Reliability: Repeated Memory Failures and False Confidence; OpenAI status page. All read October 11, 2026.