OpenAI retires older ChatGPT models

OpenAI has been clearing out older models from ChatGPT, and the important part is this: the company did not just hide a few options (help.openai.com)product on February 13, 2026. But the API stayed up for those families, at least for now. So if you use ChatGPT like a normal person, your choices shrank. If you build on OpenAI’s platform, the story is more complicated. (help.openai.com) ### Which models actually disappeared? The(openai.com)ni, and GPT-5 Instant plus GPT-5 Thinking. OpenAI framed this as a product cleanup — fewer old options, more focus on newer defaults. One wrinkle: one OpenAI page mentions GPT-5 Pro in the retirement notice, while several Help Center pages emphasize Instant and Thinking. The stable takeaway is that the older GPT-4o/4.1/o4-mini set is gone from ChatGPT, and GPT-5’s earlier variants were swept up in the same February transition. (openai.com)nAI’s ChatGPT retirement notice says the API had “no changes at this time.” The API deprecations page also shows a different lifecycle entirely, with shutdown dates and substitute models listed model by model. Basically, ChatGPT is being managed like a consumer app with a curated shelf, while the API is being managed like infrastructure with migration windows. (openai.com) ### What about enterprise customers? They got a short grace period, not a permanent exemption. OpenAI’s Busine(openai.com)inside Custom GPTs until April 3, 2026. Another Help Center page says that after April 3, GPT-4o would be fully retired across all plans. Since today is May 3, 2026, that bridge period is already over. (help.openai.com) ### Why would OpenAI do this? Because model menus can turn into junk drawers fast. Every extra option creates support burden(openai.com) says retiring models lets it focus on improving the ones most people use today. That sounds tidy, but it also means users who liked a specific model’s tone, speed, or reliability can lose it overnight. Frontier model companies keep selling progress, and product stability usually loses that fight. (openai.com) ### Why does (help.openai.com) favorite model gone. A developer can often keep shipping, at least for a while, because the API deprecation process is slower and more explicit. Think of it like a restaurant changing the menu while the wholesale kitchen still stocks the ingredients. The dining room changes first. The back end changes later. (openai.com) ### What does this mean for builders? It means you should not hard-wire your product to one frontier model’s per(openai.com)5.1 models were removed from ChatGPT on March 11, 2026, barely a month after the February retirements, with conversations mapped onto newer versions. If your app depends on one exact model staying put, you are building on sand. Abstraction layers, evals, and fallback routing are not nice-to-haves anymore. They are the seatbelt. (help.openai.com) ### So what’s th(openai.com)GPT model” and “OpenAI API model” are now two different kinds of product promises. One is a fast-moving interface. The other is a platform with a longer fuse. If you are just chatting, that means less control. If you are building, it means the safest assumption is churn. (openai.com)