GPT-4o's first API snapshot shuts down on October 23
Eight months after GPT-4o left ChatGPT, the people who refuse to say goodbye are counting down to a new date — and a community tool built on DeepSeek's agent harness lands this weekend.
OpenAI will deactivate gpt-4o-2024-05-13, the very first GPT-4o API snapshot, on October 23 — 12 days from now. The company's own deprecation page lists the shutdown with a suggested replacement, and it's the last remaining line item for the model that debuted in May 2024 with Altman's one-word post, "her." The Keep4o campaign that has dogged OpenAI since GPT-5's launch isn't going quietly: the Change.org petition to keep GPT-4o available on ChatGPT sits at roughly 24,000 signatures, and the movement's newest ask — raised in an October 7 post on X — is for OpenAI to publish the retired model's weights under an archival license, keeping the trained model without the training data or recipe. What matters here isn't nostalgia for an old chatbot; it's that OpenAI's stated retirement logic ("only 0.1% of users still choose GPT-4o each day") has become the standard template for how labs retire models users formed attachments to, while an academic analysis of more than 60,000 Keep4o posts shows the attachment is documented, not imagined. The 0.1% answer treats a model like a deprecating API endpoint. A growing body of users — and now a small literature — treats it like something else.
A community-built "DeepSeek-Bot" plugin is turning DeepSeek Harness into a Grok Bot-style multi-agent playground. The project, an unofficial community plugin (the README says so explicitly), shipped its first release on October 10 and has already pulled in over 500 GitHub stars, with commits landing the same day. Each bot gets its own name, role, model, and a persistent memory that lives as local Markdown files with automatic review after every turn; a supervisor "main bot" answers by default in group chats and dispatches to the others, which keeps token spend down. Everything is a plugin, data stays on your machine, and any model DSH can reach can be swapped into any bot — all under Apache 2.0. It's a small release, but it's a clean signal of where the agent-harness ecosystem is heading: personality-and-memory layers are becoming the thing communities build first, on top of whatever harness wins the star count.
What to watch: whether OpenAI answers the archival-weights request at all, or just lets October 23 pass in silence — so far, that has been the company's only reply to a year of petitions.
Should retired models get an archival-weights policy by default? Tell us in the comments.
