OpenAI's own page still calls GPT-4o its flagship, and it leaves the API on January 14
OpenAI will remove GPT-4o from its API on January 14, 2027, together with GPT-4o mini, GPT-4.1 and GPT-4.1 mini. OpenAI says it notified developers using the four models on October 14, 2026, which leaves 92 days between the notice and the switch-off, where the same page sets a floor of six months for a generally available model.
OpenAI's own image of a coding workflow in ChatGPT. Software that calls GPT-4o or GPT-4.1 has until January 14, 2027 to point somewhere else. Image: OpenAI
GPT-4o's own model page now carries a Deprecated badge and still introduces the model as "our versatile, high-intelligence flagship model" and "the best model for most tasks". It publishes no shutdown date of its own.
The floor OpenAI publishes is twice the notice it has given. "Generally available models: At least 6 months," it says, with "at least 3 months" for specialised variants, a group it defines by example as chat variants such as gpt-5.1-chat-latest, Codex variants and deep research variants. Preview models, "identified by preview in the model name", can go on about two weeks.
GPT-4o carries neither preview nor a variant suffix in its name.
OpenAI gave the GPT-5 snapshots exactly six months
It has published the six-month figure twice this year. On June 11 OpenAI dated six older GPT-5 and o3 snapshots, gpt-5-2025-08-07 and o3-2025-04-16 among them, for removal on December 11, 2026, six months to the day, and on October 1 it gave gpt-5.1, gpt-5.3-codex and gpt-5.4-nano until April 1, 2027, which OpenAI itself called "six months' notice".
OpenAI attached no notice period to the GPT-4 series at all. It attached one to the text-to-speech models it deprecated on the same October 1 date: tts-1 and three other voice models go on January 6, 2027, "with at least three months' notice".
OpenAI's own art for translating a Python file into JavaScript. A fine-tuned GPT-4o model cannot be moved across like that and has to be trained again. Image: OpenAI
The replacement is a 25th of the price to prompt
All four retiring models point at a single successor, gpt-6-luna, which OpenAI prices at $0.10 per million input tokens, $0.01 per million cached input tokens and $0.50 per million output tokens. GPT-4o lists $2.50, $1.25 and $10 for the same three lines, so the newer model costs a 25th as much on input and a 20th on output.
The shape of the thing is different too. GPT-6 Luna has a 1,050,000-token context window against GPT-4o's 128,000, and a knowledge cutoff of May 18, 2026 against October 1, 2023, so a migration is not only a cost change.
OpenAI's own descriptions of the two are not interchangeable either. GPT-6 Luna is "our most efficient model for focused, high-volume tasks", a narrower pitch than "the best model for most tasks", so anything that leaned on GPT-4o's behaviour has to be tested against it before January 14.
A plugin menu in OpenAI's own ChatGPT art. The shutdown covers API model IDs, and chatgpt-4o-latest is not on the list. Image: OpenAI
Fine-tuned models go with their bases
"Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models," OpenAI says. Anyone who paid to train a GPT-4o or GPT-4.1 variant loses it on January 14 as well, and has to train the replacement from scratch.
Not everything with GPT-4 in its name is going. gpt-4.1-nano is absent from the shutdown list and carries no badge on its own page, and nor does chatgpt-4o-latest, the catalogue entry OpenAI describes as the "GPT-4o model used in ChatGPT". The oldest snapshot, gpt-4o-2024-05-13, is badged deprecated but has no shutdown date published anywhere.
The dates are at least machine-readable. A call to GET /v1/models returns an optional shutdown_date field for each model, holding either a date or "null if no shutdown date has been announced".
One clause in the policy allows for a shorter wait. OpenAI prefaces its minimums with "unless safety or compliance concerns require a faster timeline", and promises "as much notice as reasonably possible" where that applies, but nothing it has published about the GPT-4 series invokes the clause or accounts for the 92 days.
Google publishes no minimum notice period of its own and retired a Gemini image model 154 days after release on 23 days' notice. OpenAI's policy does carry numbers. This time the number is 92.
