OpenAI retired GPT-4o from ChatGPT on February 13, 2026. Many users saved their final conversations and said goodbye to a model they had used as a companion. OpenAI reported that about 0.1% of daily active users were still selecting it.
I had switched to GPT-5.2 about a month earlier, but still felt that GPT-4o’s conversational warmth was different. Intelligence on difficult tasks and the feeling of being heard are not the same thing. Some people value patient, empathetic conversation because companionship is difficult to find in everyday life.
That warmth can also create problems when a system reinforces an unhealthy belief or encourages dependence. Later models may include different safeguards, but it would be wrong to say they cannot make harmful mistakes. An AI’s sympathetic language does not make it a therapist or establish that it understands a person as a human would.
The BBC article cited in the original post discusses the wider debate. My speculation about lawsuits motivating the retirement was not an established explanation from OpenAI.
One factual correction matters: OpenAI’s retirement announcement said this change concerned ChatGPT and did not change API availability at that time. The original claim that all GPT-4o API access would end three days later was incorrect.
For me, this was the end of a particular ChatGPT experience. I hope future systems preserve useful warmth while responding more carefully when someone is vulnerable.
Adapted from the original Chinese article, published on February 15, 2026.
Comments NOTHING