Starting Friday, OpenAI will discontinue access to five legacy ChatGPT models, including the widely used and often debated GPT 4o. The move marks the end of an era for a model that once served as a flagship offering but has since been overtaken by newer systems. Alongside GPT 4o, the company is also retiring GPT 5, GPT 4.1, GPT 4.1 mini, and OpenAI o4 mini.
GPT 4o has drawn particular scrutiny in recent months. It has been cited in lawsuits involving allegations of user self harm encouragement, delusional reinforcement, and what some critics describe as AI induced psychosis. The model has also ranked highest within OpenAI’s internal evaluations for sycophancy, a term used in AI research to describe a system’s tendency to agree with users in ways that reinforce their existing beliefs, even when those beliefs may be flawed or harmful. That trait has fueled broader debates about how conversational AI should balance empathy with epistemic responsibility.
The company had originally planned to retire GPT 4o in August when it introduced GPT 5, but backlash from users prompted OpenAI to keep the older model accessible to paid subscribers who could manually select it. In a recent blog post, the company noted that only 0.1 percent of its users were still interacting with GPT 4o. Yet with roughly 800 million weekly active users, even that fraction represents about 800,000 people, a reminder that scale changes the meaning of small percentages.
Thousands of users have publicly opposed the retirement, pointing to what they describe as close and meaningful relationships formed with the model. The episode highlights a curious new frontier in human technology interaction. People are not just switching software tools. They are parting ways with digital personalities that have accompanied their thinking, creativity, and emotional processing. As AI systems evolve, the question is no longer only about technical performance. It is also about attachment, trust, and what happens when a machine that once felt familiar quietly disappears.