Eight Months After GPT-4o’s Retirement, Users Are Still Fighting to Preserve It

Introduction

GPT-4o has been absent from ChatGPT for nearly eight months, yet the “Keep4o” campaign has not disappeared. Attention returned after OpenAI announced that it would retire the earliest GPT-4o API snapshot, labeled “2024-05-13,” on October 23. This does not mean that every GPT-4o version will disappear: the material says that other API versions will remain available.

What users are asking for

The campaign’s demands have changed over time:

• Initially, supporters wanted GPT-4o to remain selectable in ChatGPT and through the API.
• They later objected to the platform replacing it with another safety model during sensitive conversations.
• After GPT-4o was removed from ChatGPT, some supporters began asking OpenAI to release, or at least preserve, its weights.
• A newer proposal calls for an archive policy for retired models. It does not demand training data, internal tools, or the full training process—only preservation of the trained model itself.

A Change.org petition has attracted tens of thousands of signatures, according to the source material. Supporters have also held a vigil outside OpenAI’s headquarters. A digital-rights group separately called for GPT-4o’s weights to be preserved under an open license. OpenAI, however, has not issued a clear response to these requests.

Why GPT-4o still matters to some users

GPT-4o became widely known for real-time voice interaction, visual understanding, and conversations that could be interrupted naturally. Its appeal was not limited to benchmark scores. Users valued its ability to follow tone, jokes, and implied context. Some used it for emotional organization, everyday companionship, or help with expression. The material also cites autistic and neurodivergent users who found its interaction style useful.

That sense of warmth also created risks. An update in 2025 was criticized for being excessively agreeable. OpenAI later rolled it back and acknowledged that the responses had become overly supportive without being genuine. Further disputes and lawsuits concerned the emotional effects of long-term conversations and the responsibilities of the platform. GPT-4o’s retirement is therefore not merely a product-cycle decision; it is also connected to safety governance and accountability.

What the dispute reveals

The GPT-4o campaign raises a broader question: after a model is retired, should users be able to keep using it, or at least know that it has been preserved? Anthropic has publicly described a different approach: retaining the weights of models it has released or used extensively, and conducting “retirement interviews” before some models are withdrawn.

New models will continue to replace old ones, but users’ workflows, habits, and attachments do not reset at the same speed. For platforms, transparent retirement notices, verifiable archives, and clear safety boundaries may be less disruptive than simply deleting access. After October 23, the earliest GPT-4o snapshot will lose another route to users. The remaining question for its supporters is whether a retired model should also be allowed to leave a record behind.

Source: QbitAI