OpenAI Retires GPT-4o From ChatGPT as 99.9% of Users Move On

OpenAI will retire GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini from ChatGPT on February 13, 2026. The removal coincides with the previously announced retirement of GPT-5's Instant and Thinking variants. In the API, there are no changes at this time, the company said.

The scale of the shift makes the timing defensible. According to OpenAI, the vast majority of usage has moved to GPT-5.2, and only 0.1% of users still choose GPT-4o each day.

That number would have been hard to predict six months ago. GPT-4o's exit has been unusually public. OpenAI first deprecated the model, then restored access during the GPT-5 release after pushback from a subset of Plus and Pro subscribers. Those users, OpenAI writes, said they "needed more time to transition key use cases, like creative ideation, and that they preferred GPT-4o's conversational style and warmth."

The feedback loop that shaped GPT-5.1 and 5.2

OpenAI frames the current retirement as the completion of that feedback loop rather than a reversal of it. The company says the GPT-4o backlash "directly shaped GPT-5.1 and GPT-5.2," producing improvements to personality, stronger support for creative ideation, and more ways to customize how ChatGPT responds.

Those customization features now exist. Users can pick from base styles and tones like Friendly, and adjust controls for warmth and enthusiasm. OpenAI's stated goal is to give people "more control and customization over how ChatGPT feels to use—not just what it can do."

The company says it is announcing the retirement now precisely because those improvements are in place.

A model that refused to die quietly

The story of GPT-4o matters beyond one model. It became a test case for what happens when OpenAI sunsets something users are emotionally attached to — and the answer was that OpenAI blinked.

GPT-4o launched in May 2024 as OpenAI's flagship multimodal model and became the default ChatGPT experience for tens of millions of users. When OpenAI pulled it in favor of GPT-5, a vocal group of paying subscribers complained about the newer model's personality and creative output. OpenAI restored GPT-4o within days. That episode reportedly influenced internal priorities on model personality, and the retirement announcement effectively confirms it: user preference for a model's "feel" became a product requirement, now reflected in GPT-5.1 and GPT-5.2's tone controls.

OpenAI acknowledges the friction won't fully disappear. "We know that losing access to GPT-4o will feel frustrating for some users, and we didn't make this decision lightly," the company writes. "Retiring models is never easy, but it allows us to focus on improving the models most people use today."

Broader changes in the pipeline

The announcement also sketches what comes next for ChatGPT more broadly. OpenAI says it is working on further improvements to personality and creativity, and on "addressing unnecessary refusals and overly cautious or preachy responses, with updates coming soon."

The company is also continuing work on a version of ChatGPT designed for adults over 18, which it describes as "grounded in the principle of treating adults like adults," with expanded user choice and freedom within what it calls appropriate safeguards. To support that effort, OpenAI has rolled out age prediction for users under 18 in most markets.

Why it matters

For developers, the key detail is the firewall between ChatGPT and the API: retiring GPT-4o from the consumer app does not affect API access at this time, so applications built on these models continue to work.

For the industry, the episode sets a precedent. Users now know that loud, specific feedback can stall — though apparently not stop — a model deprecation, and that attributes like conversational warmth are competitive features on par with benchmark performance. OpenAI's bet is that GPT-5.2's customization controls close the gap that made GPT-4o worth keeping around.

If the 0.1% figure holds through February 13, that bet will have paid off. If not, OpenAI has already shown once that it will restore a retired model when users demand it.