“Chat memory full” after 10–15 messages: OpenAI Community debates the cause

⚡ The gist of the debate

In an OpenAI Community thread (289 comments, 6,031 views), users are dissecting the “Chat Memory Full” error that pops up in chats after just 10–20 messages. The post’s author, JoshB2084, immediately clarifies that he doesn’t know the root cause, but insists the behavior doesn’t look like a regular context-length limit. From there, participants split on the main question: is it a service-side glitch, an overloaded account memory, or a new tier boundary?

🗣 What participants are saying

JoshB2084 describes a week of observations: tests across different conversations and devices, cross-checked against OpenAI’s status history. His key argument is that the error appears even in small chats, and continuing “via a brief summary” hits the same barrier again after roughly 10–20 messages. That, he says, is exactly what raised his suspicion that the size of the conversation isn’t the issue.

A user under the handle fen.dan writes that he hit this for the first time in the past week and it never happened before: “I don’t believe it has anything to do with the platform.” Creating new conversations didn’t help him, and he lost a long chat — continuing via a summary didn’t work, with the error returning after about 10 messages.

fmara502 replies in Portuguese and confirms the same picture: at first he thought the chat’s shared memory was overloaded, but now he believes it’s a collective problem. The web version, in his observation, doesn’t help either.

architverma560 pins the start to September 30 and specifically notes that this isn’t the usual context limitation. With normal overflow the model gradually forgets earlier content; here, you have to end the conversation — including in new chats with few messages.

derann gives an even shorter window: after 5–10 ordinary messages comes an offer to pay or start a new chat, and he considers that framing absurd. brianosastro is on the Go subscription and says this error has him thinking about canceling — especially since, in his view, the service is “taking features away.”

Nullification took a measurement approach: he counted messages until the error and got around 15. A month ago, as he recalls, the mechanics were different — you had to wait 2–5 hours, but the number of messages was practically unlimited.

⚖️ Where they agree, where they diverge

There is a common denominator: for several participants the error reproduces in short chats, new conversations and the web client don’t fix it, and it doesn’t look like gradual forgetting of earlier context. Something else matches too — nobody names a verified cause; JoshB2084 explicitly declines the role of someone who knows the root cause.

They diverge on two points. First, the nature of the glitch: fen.dan doesn’t believe it’s a platform problem, fmara502 speaks of a collective issue, and brianosastro ties it all to the Go tier and feature cuts. Second, the threshold: 5–10 messages for derann, around 15 for Nullification, 10–20 for the post’s author. The spread is by itself a weak argument against a simple limit, but there’s no consensus on the number either.

🧪 What to check yourself

• Count messages until the error appears in a new chat, repeat 2–3 times, and compare with Nullification’s ~15 — reproducibility matters more than a single case.
• Run the same scenario in the web client and in the app: fmara502 claims the web doesn’t help, but it’s worth confirming on your own account.
• Separately test continuing “via a summary” — for fen.dan the error returned there too, after ~10 messages.
• Compare behavior on a free vs. paid account (Go is mentioned in the thread) — the tier theories are exactly where opinions diverge.
• Check OpenAI’s status page and incident history for the dates in question: the post’s author relied on it but gave no specifics in the thread.

🔗 Sources

OpenAI Community — Why is ChatGPT showing “Chat memory full” after just a few messages?

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