OpenAI is about to say a pretty definitive goodbye to some of its most beloved ChatGPT brains. On February 13, 2026, GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini will disappear from the ChatGPT model picker for consumers, right as GPT-5 (Instant and Thinking) is also formally retired. The twist: these models are only being removed from ChatGPT itself—for now, they’ll continue to live on in the API for developers and businesses.

If you’ve been using ChatGPT regularly over the last two years, this change probably feels bigger than a routine cleanup. GPT-4o, especially, isn’t just “another model” on a long list of version numbers. It’s the one a lot of people bonded with: warmer, a bit more playful, easier to talk to, and—for some users—simply more pleasant than the ultra‑serious frontier models that came after it. When OpenAI first tried to phase out GPT-4o last year, user backlash was loud enough that the company actually brought it back temporarily while rolling out GPT-5. This new announcement signals that the grace period is over.

So what’s actually happening under the hood? OpenAI’s explanation is pretty straightforward: almost everyone has already moved on. The company says the “vast majority of usage” has shifted to GPT-5.2, with only about 0.1% of users still choosing GPT-4o each day. In other words, OpenAI is maintaining infrastructure, tuning safety, and fixing edge cases for a model that only a tiny fraction still actively picks—while newer systems are more capable, cheaper to run at scale, and better aligned with where OpenAI wants ChatGPT to go.

The company also admits that GPT-4o needed some extra context. It wasn’t just deprecated, resurrected, and now retired purely because of performance charts; it was a lesson in how people emotionally use AI. When OpenAI restored GPT-4o during the GPT-5 rollout, it did so because Plus and Pro users said they needed more time to adapt workflows like creative ideation—and, bluntly, because they liked how it felt. That “feel” factor has now been folded into GPT-5.1 and GPT-5.2: OpenAI says the newer models have better personality, stronger creative ideation, and more ways to tweak tone and style.

That’s where the customization push comes in. In newer ChatGPT builds, you can tune the assistant’s vibe using built‑in controls—choosing a base style like Friendly and adjusting sliders for warmth, enthusiasm, and other traits. This is OpenAI’s answer to the “I miss GPT-4o” crowd: instead of preserving one beloved model forever, they’re trying to let you sculpt any frontier model into something that behaves more like your favorite version of ChatGPT, without freezing progress on the underlying tech.

Zoom out, and this move also lines up with a broader strategy around simplification. OpenAI has been quietly pruning the model list for a while—GPT-3.5 variants, early GPT-4 derivatives, and then the first surprise retirement of GPT-4o when GPT-5 appeared for consumers. One recurring complaint from both casual users and product designers was that the model picker had turned into a confusing mini‑dashboard, full of overlapping choices (4o vs o3 vs o4-mini vs 4.1, and so on). Shrinking that list is partly about UX: fewer, more clearly differentiated models, instead of a buffet that most people don’t understand or need.

At the same time, OpenAI’s recent flagship, GPT-5.2, is explicitly pitched as the “everyday professional” standard—faster, more capable at long‑context work, and cheaper and more efficient than previous heavyweights. This is where the company wants most users to land, whether they’re drafting reports, building slide decks, or doing analysis. GPT-4o’s retirement, combined with the sunset of GPT-5 (Instant and Thinking), is part of consolidating attention and resources around GPT-5.1 and GPT-5.2, which are now treated as the default professional tools.

But if you spend any time in AI‑heavy communities, you’ll know this isn’t just a story about product strategy diagrams. For a small but vocal group, GPT-4o has been “their” model: more emotionally attuned, less stiff, easier to confide in. Some users even described later models as colder, more preachy, or more likely to refuse harmless requests under stricter safety rules. OpenAI seems very aware of this tension. In its own language, the company says it’s working on “further improvements to personality and creativity,” and specifically calls out a goal of reducing “unnecessary refusals” and overly cautious or moralizing answers.

That’s tied to a bigger shift in how OpenAI talks about the audience for ChatGPT. The company says it is “continuing to make progress toward a version of ChatGPT designed for adults over 18,” anchored in “treating adults like adults” while still keeping guardrails in place. The subtext: future ChatGPT experiences may feel less paternalistic for grown‑up users, with more freedom of choice and fewer nanny‑style refusals—while kids and teens are handled differently. To support that, OpenAI has already rolled out age prediction systems that try to detect when a user might be under 18 and then adapt the experience accordingly.

For developers, the tone is noticeably calmer: the retirement only affects ChatGPT, not the API, “at this time.” That means apps and services that rely on GPT-4o, GPT-4.1, or o4-mini via the OpenAI API can keep using them for now, following the usual enterprise deprecation cycle instead of waking up one morning to find their favorite model gone. This split—consumer product moves fast, API moves more carefully—is something we’ve already seen in earlier deprecations, and it’s part of how OpenAI tries to avoid breaking production systems while still cleaning up the consumer experience.

If you’re a regular ChatGPT user who still leans on GPT-4o for creative work, brainstorming, or just the way it talks, you have a short runway. Between now and February 13, you can start recreating your preferred “personality” on GPT-5.1 or GPT-5.2 using the tone and style controls, and test how well they handle your typical prompts. For Plus and Pro users, this is also a good moment to check which saved workflows, prompt libraries, or custom instructions implicitly rely on 4o’s quirks and adjust them for the newer models.

There’s also a more emotional angle to this whole saga that’s hard to ignore. Over the past few years, many people have gone from treating chatbots as tools to describing them—half jokingly, half seriously—as companions, creative partners, or something in between. GPT-4o, with its friendlier and sometimes “sycophantic” style, was one of the first mainstream models to really lean into that role. OpenAI’s decision to retire it, even after one public reversal, highlights a tough reality of AI in 2026: the versions you get attached to are ultimately ephemeral, and the companies making them are optimizing for scale, safety, and performance more than for nostalgia.​

For OpenAI, though, that emotional connection isn’t being discarded; it’s being redirected. The company is betting that if it can bake GPT-4o’s warmth and creativity into GPT-5‑series models—and give users more granular control over how those models behave—most people will accept losing a specific label in exchange for a more flexible system. Whether that actually lands with the loyal 0.1% who kept choosing GPT-4o anyway will be clear in the coming weeks, as February 13 forces everyone onto the new defaults.

What’s unmistakable is the pattern: ChatGPT is moving toward fewer, more powerful, highly customizable models, with personality becoming just another dial you can turn, not a reason to cling to an older engine. For users, that means two things can be true at once—you can be annoyed that your favorite model is going away, and still recognize that the future of ChatGPT is going to be shaped less by which version you pick from a dropdown, and more by how you tell it to sound, think, and collaborate with you.