OpenAI shuts off GPT-4 and o1 Oct 23

// the October 23 list, as published

OpenAI's deprecations page introduces the list with one sentence:

Every row below reads October 23, 2026. The first name in each row is the snapshot; the names after it are the aliases that point at it. The right-hand column is OpenAI's listed substitute.

That is twelve rows and twenty-four model names. What happens on the day is stated plainly on the same page: "At the time of the shut down, the model or endpoint will no longer be accessible."

Plain names are on it too

It is easy to read a list of dated snapshots and assume the short names are safe. They are not. The plain names gpt-4, gpt-3.5-turbo, gpt-4-turbo, gpt-4.1-nano, o1, o1-pro, o3-mini and o4-mini are listed alongside their snapshots, and those are exactly the names years of tutorials and sample code hard-code.

What is not on this list: plain gpt-4o. Only its oldest snapshot, gpt-4o-2024-05-13, goes on October 23. The gpt-4o model page shows the plain name pointing at gpt-4o-2024-08-06 today ("Default snapshot: gpt-4o-2024-08-06"). The page does not say what a plain name does after its only snapshot is removed, so neither do we. And chatgpt-4o-latest is not a safe fallback: it was removed from the API on February 17, 2026.

So search for both kinds. A grep for the dated names alone misses the most common ones.

Fine-tunes go with their base

The same section adds a second table:

ft-o4-mini-2025-04-16 sits in the main table on the same date. The rule behind it is in OpenAI's fine-tuning section: "Inference on fine-tuned models will be disabled only when the underlying base model is deprecated." A fine-tuned model is tied to its base: when the base goes, so does the model you trained on it. The fine-tune table lists a recommended replacement base for each (gpt-5.6-terra, -sol or -luna).

The video shows the gpt-3.5-turbo, gpt-4, gpt-4.1-nano and o4-mini fine-tunes, whose bases go the same day. The babbage-002 and davinci-002 bases were on an earlier list (September 28, 2026).

Move, then test before the date

The fix is mechanical, and the test is not optional:

Treat the pattern above as a starting point and check what it matches: model names also live in config files, environment variables and database rows. The substitute is a different model, so the thing to check is your own app's behaviour on it, before the day the old name stops answering.

OpenAI also says: "When we announce a model deprecation, we notify customers who are actively using the model by email and document the deprecation on this page." If the email went to someone who has left, this page is the one to check.

It isn't the last date

The next GPT-5 and o3 batch is already announced: "On June 11, 2026, we notified developers using older GPT-5 and o3 model snapshots of their deprecation and removal from the API on December 11, 2026." That list includes gpt-5-2025-08-07, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07, gpt-5-pro-2025-10-06, o3-pro-2025-06-10 and o3-2025-04-16, which is the snapshot plain o3 points at today ("Default snapshot: o3-2025-04-16" on the o3 model page).

That does not make pinning the mistake. OpenAI's text guide says "we strongly recommend: Pinning your production applications to specific model snapshots … to ensure consistent behavior", and a pinned name gives you a date you can see on this page. A plain name is only a pointer to a snapshot, so either way the habit to add is the same: check the deprecations page.

Scope. This is OpenAI's own API. Azure OpenAI and other hosts publish their own retirement dates for the same model families. Other shutdowns on the same page fall on October 1, November 30 and December 1.

Related: model routing covers choosing a model per step, which is where many hard-coded names live.

Sources: OpenAI API docs — Deprecations (2026-04-22 legacy GPT model snapshots; 2026-06-11 GPT-5 and o3; fine-tuning availability; 2025-11-18 chatgpt-4o-latest), the gpt-4o and o3 model pages, and the text generation guide. Verified 2026-09-26. The page can change; check it before you act.