https://enterprise.shutterstock.com/image-photo/openai-logo-displayed-on-smartphone-screen-2520388517
or
https://enterprise.shutterstock.com/image-photo/chatgpt-logo-displayed-on-smartphone-screen-2520385879
Last August, ChatGPT developers OpenAI unceremoniously killed the fan favorite GPT-4o model, before giving in to complaints and bringing it back a week later. Now, the company’s taking a second swing at getting its users to move on. In a new post to its website, OpenAI announced that it’s retiring GPT-4o again.
The model’s set to disappear from ChatGPT’s model picker on Feb. 13, alongside other older models like GPT-4.1, GPT-4.1 mini, and OpenAI o4-Mini. And OpenAI is clearly nervous about the decision.
“While the announcement applies to several older models,” OpenAI wrote, “GPT-4o deserves special context.”
According to the company, it has taken user outcry over the initial deprecation of 4o to heart while developing its newest models, GPT-5.1 and GPT-5.2, and has built these models with the idea of maintaining the features fans liked best about the old model. The company says that now “only 0.1% of users” opt for GPT-4o on a daily basis.
As such, the company wants to focus on “improving the models most people use today,” which apparently means removing older ones. “We know that losing access to GPT-4o will feel frustrating for some users, and we didn’t make this decision lightly,” the post reads.
What’s the big deal with GPT-4o?
So, what’s with OpenAI treating its users so gingerly, especially when GPT-4o is a few generations behind, and there are newer models that supposedly do everything it does, but better?
Well, when GPT-4o was first deprecated, people weren’t happy. Users called its successor, GPT-5, “an unmitigated disaster,” and accused OpenAI of pulling “the biggest bait-and-switch in AI history.”
Some criticized the model’s usefulness, saying it got answers wrong and broke code, but what maybe stuck out the most was people calling out its more concise tone.
GPT-4o has been called “sycophantic” by critics, something the company addressed and said it wanted to pull back on in future updates. But I guess one person’s “yes man” is another person’s “active listener.” When the company initially pulled GPT-4o, users complained that its replacement was cold and felt less like a “friend.” Even OpenAI acknowledged this, saying in today’s post that users “preferred GPT-4o’s conversational style and warmth.”
In short, in the words of 4o-supporters themselves, they were “grieving” the model.
Is GPT-5.2 a good replacement for 4o?
That said, with so many users now seeming to have moved on from 4o, OpenAI’s decision does seem understandable on the surface. Personally, one of the things that drives me away from AI is how much reassuring filler text seems to fluff up most answers (“you’re absolutely right” and such), seemingly just to make me feel good about myself. More concise, to-the-point responses would be a little less off-putting for me.
What do you think so far?
To try to split the difference, OpenAI reworked its Personalization feature in GPT-5.1, so users can simply choose how the chatbot will treat them. There are options for more professional responses, more nerdy ones, more efficient ones, and for those who want that active listener style, more friendly ones.
Going by OpenAI’s numbers, that seems to have been enough for most people, but there are still some calling foul at the company’s new announcement.
GPT-4o loyalists are still out there
In a Reddit thread responding to OpenAI’s new posts, users doubted that the 0.1% number for 4o was accurate, saying that prompts have been “rerouting to 5.2 no matter what” and that “something somewhere in their calculations doesn’t add up.” Others pointed out that free users can’t use GPT-4o and that it’s not enabled by default, which will naturally juice the numbers against it.
As such, calls to cancel ChatGPT subscriptions are once again circulating amongst 4o’s more dedicated fans. In a popular thread on the OpenAI subreddit, one user called 4o “OpenAI’s most advanced and beloved model,” and praised its “personality, warmth, and consistency,” saying that its fans have built long-term project and “emotional support routines” around it, and that suddenly losing it without even the option for a legacy mode “feels abrupt and deeply disappointing.”
“This isn’t about resisting innovation,” the post writes. “It’s about respecting bonds users have formed with specific models.”
Whether the fan outcry will work again remains to be seen. However, as ChatGPT chief Nick Turley has previously looked at those kinds of bonds with skepticism, and because keeping old models in operating condition probably takes developer resources away from making new ones, I wouldn’t count on it.
