OpenAI retires models from ChatGPT
- OpenAI retired GPT‑4o, GPT‑4.1, GPT‑4.1 mini, o4‑mini, and GPT‑5 Instant/Thinking from ChatGPT on February 13, 2026 for Business, Enterprise, and Edu users. - The big tell is what did not change: those same models still remain in the API, and GPT‑4o inside Custom GPTs lasted until April 3. - That splits ChatGPT from the API more clearly — consumer-style defaults in one place, stable building blocks for teams in another.
OpenAI has been quietly drawing a harder line between ChatGPT the product and OpenAI’s API the platform. The clearest sign is model retirements inside ChatGPT itself. On February 13, 2026, OpenAI removed GPT‑4o, GPT‑4.1, GPT‑4.1 mini, o4‑mini, and GPT‑5 Instant and Thinking from ChatGPT for Business, Enterprise, and Edu users — while leaving API access alone. (help.openai.com) ### What actually got removed? Inside ChatGPT workspaces, those older model options are gone from the picker. This applied across Business, Enterprise, and Edu, and OpenAI repeated the same language across its help pages for each tier. There was one short grace period — GPT‑4o stayed available inside Custom GPTs until April 3, 2026, even after disappearing from normal ChatGPT use. (help.openai.com) ### What did not change? The API. That is the important part. OpenAI’s retirement notice says the ChatGPT removals do not affect API availability “at this time,” which means developers can still build on those models even if ChatGPT users inside managed workspaces cannot click into them anymore. Basically, the chat app is being treated as a curated surface, not a promise that every old model stays selectable forever. (help.openai.com) ### Why does that matter? Because a lot of companies blurred the two together. If a model showed up in ChatGPT, teams assumed it was part of the stable toolkit. Turns out OpenAI is signaling the opposite. ChatGPT can change fast — defaults, personalities, model menus, all of it. The API is the steadier layer where developers are suppos(help.openai.com) it is the clear implication of the split. (help.openai.com) ### Why were some users upset about 4o? GPT‑4o had fans for reasons that were not just benchmark scores. Some users liked its tone, speed, and what they described as more personality. That is why the retirement landed as more than a housekeeping move. Business Insider’s recent write-up on GPT‑5.5 framed it as a partial return of that older “spark,” which tells you the complaint was about feel as much as capability. (africa.businessinsider.com) ### So is GPT‑5.5 the replacement? In practice, yes for many ChatGPT users — but not in a neat one-to-one way. OpenAI’s current help pages point users toward newer GPT‑5.x options in ChatGPT, which suggests the company wants fewer legacy branches and more movement onto a smaller set of current models. The catc(africa.businessinsider.com)consistency, and now they may need the API to keep that control. (help.openai.com) ### What does this change for teams? It pushes them toward grown-up model management. Instead of trusting the ChatGPT picker, teams may need routing, evals, and fallback plans — one model for drafting, another for coding, another for support, maybe an older API model for a workflow that breaks when tone changes. That is more work, but it is also more (help.openai.com)updates constantly, while serious users build against the underlying platform. (help.openai.com) ### Bottom line This is not just a cleanup of old model names. It is OpenAI telling customers where stability lives. If you want whatever ChatGPT thinks is best this month, stay in the app. If you need repeatable behavior, pin your models in the API. (help.openai.com)