In an OpenAI Community thread, user Arwen_Undomiel claims that on Thursday evening, November 12, GPT-6 in Chat was working great — and on Friday morning, November 13, it showed a sharp degradation on the same prompt in High and Medium modes. The thread has racked up 38 replies, 73 likes, and 803 views. Two things are being debated: whether the regression is real, and whether it's tied to a quiet rework of Chat modeled on Work.
Arwen_Undomiel (the thread starter) describes the experience of a Plus subscriber: they used GPT-6 "many times per day," and until the evening of November 12 it was noticeably better than 5.6. On the morning of November 13, the difference on the very same prompt was so stark that the author moved some tasks back to 5.6 and is asking OpenAI to roll back the changes.
In a follow-up post, Arwen_Undomiel refines the hypothesis: Chat's output started to resemble Work's style and functions, and Chat suddenly took a liking to "commands" it never used before Friday. Hence the plea not to "nuke the chat function" and the remark: "Not all of us want to code."
tamaradenmeade confirms the observation and narrows it down: Instant stayed conversational, while High and Medium on 6.0 lost their personality and conversational tone, in her words. She also notes the model had only just started to get good: "6.0 had actually finally started to come good and now we're worse than before." She suspects she may have been placed in a test group without being told.
norcalandrew steers the conversation toward practical ground: almost nobody in these threads has filed a formal ticket through help.openai.com. He did — and got a reply saying support knows of no active incident matching this behavior.
The discussion then turns to how to even frame a complaint. tamaradenmeade admits she hasn't filed one: there's no actual breakage to point to, other than a change in voice and personality. She adds that she doesn't want the metrics to show growth in Instant usage that wasn't chosen by preference. Arwen_Undomiel and xiasiwei have both already been through support: the former managed to get past the bot and escalate the issue, the latter submitted a request and is waiting for a human response.
Arwen_Undomiel's method is simple but concrete: a saved Thursday output as the baseline, a comparison with Friday's, with the focus on Medium and High while Instant behaves normally. It's the only thing resembling a test in the thread.
They agree on the facts: the changes were noticed after the night of November 12–13, they affect High and Medium, and Instant is mostly untouched. This is about style, structure, and the character of responses — not a drop on benchmarks. Every participant mentions OpenAI support in one way or another, meaning the issue has already moved beyond "it feels off to me."
They split on the cause and the status of the problem. Arwen_Undomiel builds a case for a Chat–Work merger and a silent rollout; tamaradenmeade suspects she's in a test without notification; norcalandrew brings in support's answer that there's no incident. There's no consensus: to some participants it's a regression, to others it's a change in voice that's hard to file as a ticket — and the two readings remain unreconciled in the thread.
OpenAI Community — ChatGPT 6 Massive Output Degradation Starting Friday 11/13
Disclaimer: this is a summary of a discussion, not facts verified by our editors. The opinions belong to the participants; the numbers belong to whoever cited them.
Source: community.openai.com · post in Telegram