OpenAI reshuffles ChatGPT stack
OpenAI has retired several public ChatGPT models (GPT‑4o, GPT‑4.1, GPT‑4.1 mini, o4‑mini and GPT‑5 variants) from its consumer ChatGPT product while preserving access to GPT‑4o for Business, Enterprise and Education customers inside Custom GPTs. Separately, there are industry rumours that a new internal model codenamed “Spud” completed pretraining in late March and could ship soon. (help.openai.com) (abhs.in)
OpenAI has cleared out much of ChatGPT’s public model menu, retiring GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and earlier GPT-5 variants from the consumer product on February 13, 2026. (help.openai.com) OpenAI said the change applies to ChatGPT, not its application programming interface, so developers can still use those older models through OpenAI’s developer platform. OpenAI also retired GPT-5.1 Instant, GPT-5.1 Thinking, and GPT-5.1 Pro from ChatGPT on March 11, 2026. (openai.com) (help.openai.com) The company gave ChatGPT Business, Enterprise, and Education customers a short grace period for GPT-4o inside Custom GPTs, then ended that exception on April 3, 2026. OpenAI’s help pages now say GPT-4o is fully retired across all ChatGPT plans. (help.openai.com 1) (help.openai.com 2) That leaves ChatGPT with a tighter stack centered on newer GPT-5.3 and GPT-5.4 options, while older models stay available in the application programming interface for customers who built around them. OpenAI said retiring models lets it focus on the models “most people use today.” (help.openai.com) (openai.com) The cleanup also closes a chapter for GPT-4o, which OpenAI had positioned as its flagship “omni” model in 2024 for text, image, and voice work inside ChatGPT. GPT-4.1, by contrast, was introduced in April 2025 as an application programming interface model family aimed at coding, instruction following, and long context. (openai.com 1) (openai.com 2) Separately, reports in late March said OpenAI had finished pretraining a new internal model codenamed “Spud,” with public release discussed as a matter of weeks rather than months. Pretraining is the first large-scale pass where a model learns patterns from raw data before later tuning and safety testing. (the-decoder.com) (abhs.in) OpenAI has not publicly announced a consumer launch for Spud or confirmed what name it would ship under, and outside reports disagree on whether it would be branded as GPT-5.5 or GPT-6. The reporting tied to the codename traces to outside outlets summarizing an internal memo rather than an OpenAI product post. (the-decoder.com) (abhs.in) For ChatGPT users, the immediate change is simpler than the rumor cycle: fewer legacy choices in the app, newer default models in their place, and no remaining GPT-4o carve-out after April 3. The next public clue will be an OpenAI release note or product post, not the retired model picker. (help.openai.com 1) (help.openai.com 2)