ChatGPT's Context Window Is Filling Up Twice as Fast: Bug or a Quiet Cut

⚡ The gist of the dispute

In an OpenAI Community thread (31 replies, 105 likes, 2,842 views), users are discussing why ChatGPT now hits the context limit noticeably earlier. Thread author DoraH noticed that a conversation used to fit roughly 20 messages, but now the warning appears after ~10. There are also separate complaints that the model picks up older context worse and doesn't always handle the handoff to a new chat.

🗣 What participants are saying

DoraH insists this isn't about "Saved Memory" or the "Chat memory full" notification, but about the context window itself. In his observation, over the past few days the amount of dialogue before the limit kicks in has shrunk by roughly half, which breaks long creative projects. His second point is that ChatGPT's ability to correctly retrieve and use information from older conversations has degraded.

lorranyfernandes428 confirms: "My chat has been like this for days too," and adds that a new chat doesn't help — "the same mess." According to her, this almost never happened before September of last year, and a similar previous outage occurred on October 20. She writes that frequent breakdowns hit new users too, not just those running large projects.

ryouk broadens the picture: the limit kicks in "even in completely new, empty conversations," meaning it's not about accumulated context or history transfer. briandowmobile pins down the dates: his problem started on November 4–5 and was still ongoing on November 11, and it "also affects completely new, short chats" — with no document uploads or heavy analysis. He also notes the issue reproduces across different projects.

geoff.widders points to another symptom: previously you could ask a "frozen" chat to leave a handoff for a new conversation, but now it simply cuts off — "it just stops. End of." Slickepinde says the problem has lasted about two weeks and emphasizes that even in plain text mode, with no images or web search, the conversation cuts off quickly.

skye.daniel20009 describes the dynamics on the free plan: three weeks ago the window filled up after about 30 messages, now after 10. He wrote to OpenAI support, but "felt like I made no progress": they promised to escalate, yet the behavior didn't change. Hence his question — is this a bug or a new rule.

ThatBoringDeveloper doubts it's intentional: as he recalls, the problem seemed to be resolved by the end of October, so apart from a server-side error he sees no other cause — "at least I hope it isn't" an experiment or a deliberate restriction.

⚖️ Where they agree and where they diverge

The common denominator is simple: at least seven people reproduce the symptom, for all of them it appeared in recent days or weeks, and for all of it affects short new chats, not just long ones. Nobody disputes that the behavior has changed — they differ on the nature of the change and on whether it's "their" problem or a general one.

The real point of conflict is whether it's a bug, an experiment, or a quiet limit cut. ThatBoringDeveloper leans toward a server-side error, skye.daniel20009 poses the question outright as "bug or new rule," while lorranyfernandes428 and Slickepinde call what's happening a "grade-down." There's no consensus in the thread: no response from OpenAI is visible in the comments, and nobody cites any official confirmation of the limits either.

🧪 What to check yourself

1. Measure the threshold: start an empty chat and count the messages until the "Context window full" banner appears — compare with your past values (the thread mentions ~20 and ~10; one participant had 30 and 10).
2. Repeat the test in different projects and on a free/paid account, as briandowmobile did, to separate a systemic outage from a local one.
3. Check the "handoff": ask the chat to leave a summary before it cuts off — for geoff.widders this stopped working, so it's worth verifying whether it reproduces for you.
4. Remove everything extraneous: plain text with no files, images, or search — that's how Slickepinde tested it, and it's the cleanest way to see actual context consumption.
5. If it reproduces, write to support with dates and chat IDs: in skye.daniel20009's case, the appeal alone produced no results.

🔗 Sources

OpenAI Community — ChatGPT Context Window Filling Up Significantly Faster and Increasing Continuity Problems – Affecting Free and Paid User

Disclaimer: this is a summary of a discussion, not facts verified by the editorial team. The opinions belong to the participants, and the figures to those who provided them.