Deprecation notice: upcoming model shutdowns in 2026

API Deprecations
1

This info has just been shared via email by OpenAI. It can also be accessed via the deprecations page in the docs.

Correction: text-embedding-3-small is not being deprecated

Over the next three to six months, we will deprecate a set of older OpenAI models to improve reliability and simplify model selection, with recommended newer alternatives available for each model.

Access to these models will be shut down in the API on the following dates:

Shutdown on July 23, 2026

  • computer-use-preview-2025-03-11
  • gpt-4o-audio-preview-2024-12-17
  • gpt-4o-mini-audio-preview-2024-12-17
  • gpt-4o-mini-realtime-preview-2024-12-17
  • gpt-4o-mini-search-preview-2025-03-11
  • gpt-4o-mini-tts-2025-03-20
  • gpt-4o-search-preview-2025-03-11
  • gpt-5-chat-latest
  • gpt-5-codex
  • gpt-5.1-chat-latest
  • gpt-5.1-codex
  • gpt-5.1-codex-max
  • gpt-5.2-codex
  • gpt-5.1-codex-mini
  • gpt-audio-mini-2025-10-06
  • gpt-realtime-mini-2025-10-06
  • o3-deep-research-2025-06-26
  • o4-mini-deep-research-2025-06-26

Shutdown on October 23, 2026

  • gpt-3.5-turbo-0125
  • gpt-4-0613
  • gpt-4-1106-preview
  • gpt-4-turbo
  • gpt-4.1-nano
  • gpt-4o-2024-05-13
  • gpt-image-1
  • o1-2024-12-17
  • o1-pro-2025-03-19
  • o3-mini-2025-01-31o4-mini-2025-04-16

Finetuned versions of models being deprecated:

  • Finetuned versions of gpt-3.5-turbo
  • Finetuned versions of gpt-4.1-nano-2025-04-14
  • Finetuned versions of babbage-002
  • Finetuned versions of davinci-002
  • Finetuned versions of o4-mini-2025-04-16
  • Finetuned versions of gpt-4

We encourage you to explore our latest flagship models featured here, which offer improved performance and broader feature support.

Thanks for building with OpenAI. If you have any questions about model migration, feel free to reach out on the OpenAI Developer Forum.

2

This is an unprecendented massive list of models going away.

  • ISSUE: gpt-4-turbo appears in the list, but it is an alias to gpt-4-turbo-2024-04-09 that does NOT appear. Clarification must be made if the destination model will be shut off.

gpt-4-1106-preview appears in the listing with a shutoff date of 2026-10-23, but it was already listed and shutoff 2026-03-26, along with gpt-4-0125-preview, and is no longer available on API.

It seems that the embeddings model has been removed from the listing. If in error, the deprecations page should acknowledge the error blasted by email. Embeddings should be long-life, with a year notice.

Alias pointers should be distinct in their deprecation listing. “This pointer to an actual model is also going away”. Currently it seems we are forced to merely assume the destination model is deprecated.

Here’s a truth table for programmatic use (without the truth absolutely verified due to ambiguities), with Python inline comments. I included the image and audio models from the newest announcement, previous to that is only chat models.

DEPRECATIONS = {
    "gpt-3.5-turbo-0125": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23"},  # official ft-gpt-3.5-turbo shutoff 2026-10-23
    "gpt-4-0613": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23"},  # official ft-gpt-4 shutoff 2026-10-23
    "gpt-4-1106-preview": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23"},  # ft supported; official ft-gpt-4 shutoff 2026-10-23
    "gpt-4-turbo-2024-04-09": {"shutoff": "2026-10-23", "alias": "gpt-4-turbo"},  # announcement lists alias "gpt-4-turbo"
    "gpt-4.1-nano-2025-04-14": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23", "alias": "gpt-4.1-nano"},  # announcement lists alias "gpt-4.1-nano"; official ft-gpt-4.1-nano-2025-04-14 shutoff 2026-10-23
    "gpt-4o-2024-05-13": {"shutoff": "2026-10-23"},
    "gpt-image-1": {"shutoff": "2026-10-23"},
    "o1-2024-12-17": {"shutoff": "2026-10-23"},
    "o1-pro-2025-03-19": {"shutoff": "2026-10-23"},
    "o3-mini-2025-01-31": {"shutoff": "2026-10-23"},
    "o4-mini-2025-04-16": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23"},  # official ft-o4-mini-2025-04-16 shutoff 2026-10-23

    "gpt-3.5-turbo-instruct": {"shutoff": "2026-09-28"},  # not known to be offered for API fine-tuning
    "babbage-002": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},  # new FT training disabled 2024-10-28; official ft-babbage-002 shutoff 2026-10-23
    "davinci-002": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},  # new FT training disabled 2024-10-28; official ft-davinci-002 shutoff 2026-10-23
    "gpt-3.5-turbo-1106": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},  # official ft-gpt-3.5-turbo shutoff 2026-10-23

    "computer-use-preview-2025-03-11": {"shutoff": "2026-07-23"},
    "gpt-4o-audio-preview-2024-12-17": {"shutoff": "2026-07-23"},
    "gpt-4o-mini-audio-preview-2024-12-17": {"shutoff": "2026-07-23"},
    "gpt-4o-mini-realtime-preview-2024-12-17": {"shutoff": "2026-07-23"}
    "gpt-4o-mini-search-preview-2025-03-11": {"shutoff": "2026-07-23"},
    "gpt-4o-mini-tts-2025-03-20": {"shutoff": "2026-07-23"},
    "gpt-4o-search-preview-2025-03-11": {"shutoff": "2026-07-23"},
    "gpt-5-chat-latest": {"shutoff": "2026-07-23"},
    "gpt-5-codex": {"shutoff": "2026-07-23"},
    "gpt-5.1-chat-latest": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex-max": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex-mini": {"shutoff": "2026-07-23"},
    "gpt-5.2-codex": {"shutoff": "2026-07-23"},
    "gpt-audio-mini-2025-10-06": {"shutoff": "2026-07-23"},
    "gpt-realtime-mini-2025-10-06": {"shutoff": "2026-07-23"},
    "o3-deep-research-2025-06-26": {"shutoff": "2026-07-23"},
    "o4-mini-deep-research-2025-06-26": {"shutoff": "2026-07-23"},

    "gpt-4-0314": {"shutoff": "2026-03-26"},
    "gpt-4-0125-preview": {"shutoff": "2026-03-26", "alias": "gpt-4-turbo-preview"},  # maybe not offered for ft?

    "gpt-4o-audio-preview-2025-06-03": {"shutoff": "2026-03-24"},
    "gpt-4o-mini-audio-preview": {"shutoff": "2026-03-24"},

    "chatgpt-4o-latest": {"shutoff": "2026-02-17"},

    "codex-mini-latest": {"shutoff": "2026-01-16"},  # still working 2026-01-20

    "o1-mini-2024-09-12": {"shutoff": "2025-10-27", "alias": "o1-mini"},
    "gpt-4o-audio-preview-2024-10-01": {"shutoff": "2025-10-10"},
    "o1-preview-2024-09-12": {"shutoff": "2025-07-28", "alias": "o1-preview"},
    "gpt-4.5-preview": {"shutoff": "2025-07-14"},

    "gpt-4-32k-0613": {"shutoff": "2025-06-06", "alias": "gpt-4-32k"},  # deprecation table separately lists "gpt-4-32k"
    "gpt-4-32k-0314": {"shutoff": "2025-06-06"},

    "gpt-4-1106-vision-preview": {"shutoff": "2024-12-06", "alias": "gpt-4-vision-preview"},

    "gpt-3.5-turbo-0613": {"shutoff": "2024-09-13", "ft_shutoff": "2026-10-23"},  # earlier docs exempted FT models; official ft-gpt-3.5-turbo shutoff 2026-10-23
    "gpt-3.5-turbo-16k-0613": {"shutoff": "2024-09-13", "ft_shutoff": "2026-10-23"},  # earlier docs exempted FT models; official ft-gpt-3.5-turbo shutoff 2026-10-23
    "gpt-3.5-turbo-0301": {"shutoff": "2024-09-13"},
}

These seem to be aliases that will have no destination, or that were listed instead of the true model:

computer-use-preview, gpt-3.5-turbo, gpt-4, gpt-4-32k, gpt-4-turbo, gpt-4-turbo-preview, gpt-4-vision-preview, gpt-4.1-nano, gpt-4o-audio-preview, gpt-4o-mini-search-preview, gpt-4o-search-preview, o1, o1-mini, o1-preview, o1-pro, o3-deep-research, o3-mini, o4-mini, o4-mini-deep-research

Prominently: Killing gpt-3.5, gpt-4, and gpt-4-turbo lines completely, along with original gpt-4o, then cutting right into reasoning models like o4-mini. The only thing spared is o3.

3

This is terrible news. I strongly urge you not to shut down these legacy models, especially the finetuned ones that many of us spent disproportionate amounts of time tweaking to our needs.

You are shutting down models that countless people have relied on and depended on for years all for the sake of “improved reliability and simplified model selection”? This is simply not a good enough reason and achieves the exact opposite in reality.

The dumbing down of AI continues…

4

This is a ridiculous list of models and super fast. Seems to me this will make me move to Mistral, Chinese or OS stable models first than doing finetuning again with OpenAI. Ridiculous.

5

Indeed, these are bad news.

Particularly, what worries me about is gpt-4o-mini-tts-2025-03-20, which has not been fixed, and the replacement doesn’t make any sense at all. What does realtime has to do with tts?

image
image851Ă—298 16.2 KB

There have been complaints about the newer snapshot gpt-4o-mini-tts-2025-12-15 not properly following instructions, in which cases the older model is still necessary:

6

Do not depreciate gpt-4o-mini-tts-2025-03-20! gpt-4o-mini-tts-2025-12-15 is still broken!

7

I noticed that gpt-4o-mini-tts-2025-03-20 is not on the depreciation list anymore, Deprecations | OpenAI API

Does that mean it’s staying?

8

Thanks for raising this observation.
I’ll ask the team about it.

9

Unfortunately, it seems not. :confused:

image
image900Ă—334 18.9 KB

10

It’s back!
Must have been an error.

11

Dang that wasn’t there when I posted. They must have put it back on the list. It’s terrible. I am going with a different tts provider.

I already found one that does a great job with expressiveness.

12

I wrote a longer post about the broader user-side impact here:Restore 5.1 in ChatGPT: Replacement Is Not Migration - OpenAI Is Breaking Power-User Trust

My concern is not only this specific shutdown, but the repeated loss of stable fallback paths for power users — first in the ChatGPT UI, and now potentially in the API as well.

For some users, 5.1 is not just an old model. It is a proven assistant behavior that newer models have not fully preserved.

15

OpenAI has many more model deprecations since this post.

OpenAI documentation has blocks by announcement date.

I have them updated, and by order of shutoff date, in Python format, easily adapted to other languages by removing comments: pretty much all models of API concern including those now dead except by fine-tuning.

You can start with gpt-5 being shutoff before the end of the year, and scroll your cursor to those being killed in under two weeks.

DEPRECATIONS: dict[str, dict[str, str]] = {
    "gpt-5-2025-08-07": {"shutoff": "2026-12-11", "alias": "gpt-5"},
    "gpt-5-mini-2025-08-07": {"shutoff": "2026-12-11", "alias": "gpt-5-mini"},
    "gpt-5-nano-2025-08-07": {"shutoff": "2026-12-11", "alias": "gpt-5-nano"},
    "gpt-5-pro-2025-10-06": {"shutoff": "2026-12-11", "alias": "gpt-5-pro"},
    "o3-2025-04-16": {"shutoff": "2026-12-11", "alias": "o3"},
    "o3-pro-2025-06-10": {"shutoff": "2026-12-11", "alias": "o3-pro"},
    "gpt-image-1-mini": {"shutoff": "2026-12-01"},
    "gpt-image-1.5": {"shutoff": "2026-12-01"},
    "chatgpt-image-latest": {"shutoff": "2026-12-01"},
    "gpt-3.5-turbo-0125": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23","alias": "gpt-3.5-turbo"},
    "gpt-4-0613": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23", "alias": "gpt-4"},
    "gpt-4-turbo-2024-04-09": {"shutoff": "2026-10-23", "alias": "gpt-4-turbo"},
    "gpt-4.1-nano-2025-04-14": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23", "alias": "gpt-4.1-nano"},
    "gpt-4o-2024-05-13": {"shutoff": "2026-10-23"},
    "gpt-image-1": {"shutoff": "2026-10-23"},
    "o1-2024-12-17": {"shutoff": "2026-10-23", "alias": "o1"},
    "o1-pro-2025-03-19": {"shutoff": "2026-10-23", "alias": "o1-pro"},
    "o3-mini-2025-01-31": {"shutoff": "2026-10-23", "alias": "o3-mini"},
    "o4-mini-2025-04-16": {"shutoff": "2026-10-23", "ft_shutoff": "2026-10-23", "alias": "o4-mini"},
    "gpt-3.5-turbo-instruct": {"shutoff": "2026-09-28"},
    "babbage-002": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},  # new FT training disabled 2024-10-28
    "davinci-002": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},  # new FT training disabled 2024-10-28
    "gpt-3.5-turbo-1106": {"shutoff": "2026-09-28", "ft_shutoff": "2026-10-23"},
    "gpt-5.3-chat-latest": {"shutoff": "2026-08-10"},
    "gpt-5.2-chat-latest": {"shutoff": "2026-08-10"},
    "computer-use-preview-2025-03-11": {"shutoff": "2026-07-23", "alias": "computer-use-preview"},
    "gpt-4o-search-preview-2025-03-11": {"shutoff": "2026-07-23", "alias": "gpt-4o-search-preview"},
    "gpt-4o-mini-search-preview-2025-03-11": {"shutoff": "2026-07-23", "alias": "gpt-4o-mini-search-preview"},
    "gpt-4o-mini-tts-2025-03-20": {"shutoff": "2026-07-23"},
    "gpt-5-chat-latest": {"shutoff": "2026-07-23"},
    "gpt-5-codex": {"shutoff": "2026-07-23"},
    "gpt-5.1-chat-latest": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex-max": {"shutoff": "2026-07-23"},
    "gpt-5.1-codex-mini": {"shutoff": "2026-07-23"},
    "gpt-audio-mini-2025-10-06": {"shutoff": "2026-07-23"},
    "gpt-realtime-mini-2025-10-06": {"shutoff": "2026-07-23"},
    "o3-deep-research-2025-06-26": {"shutoff": "2026-07-23", "alias": "o3-deep-research"},
    "o4-mini-deep-research-2025-06-26": {"shutoff": "2026-07-23", "alias": "o4-mini-deep-research"},
    "gpt-5.2-codex": {"shutoff": "2026-07-23"},
    # shutoff as of 2026-07-13
    "dall-e-2": {"shutoff": "2026-05-12"},
    "dall-e-3": {"shutoff": "2026-05-12"},
    "gpt-4o-realtime-preview-2025-06-03": {"shutoff": "2026-05-07", "alias": "gpt-4o-realtime-preview"},
    "gpt-4o-realtime-preview-2024-12-17": {"shutoff": "2026-05-07"},
    "gpt-4o-mini-realtime-preview": {"shutoff": "2026-05-07", "alias": "gpt-4o-mini-realtime-preview"},
    "gpt-4o-audio-preview-2025-06-03": {"shutoff": "2026-05-07", "alias": "gpt-4o-audio-preview"},
    "gpt-4o-mini-audio-preview": {"shutoff": "2026-05-07", "alias": "gpt-4o-mini-audio-preview"},
    "gpt-4-0314": {"shutoff": "2026-03-26"},
    "gpt-4-1106-preview": {"shutoff": "2026-03-26", "ft_shutoff": "2026-10-23"},
    "gpt-4-0125-preview": {"shutoff": "2026-03-26", "alias": "gpt-4-turbo-preview"},
    "chatgpt-4o-latest": {"shutoff": "2026-02-17"},
    "codex-mini-latest": {"shutoff": "2026-02-12"},
    "o1-mini-2024-09-12": {"shutoff": "2025-10-27", "alias": "o1-mini"},
    "gpt-4o-audio-preview-2024-10-01": {"shutoff": "2025-10-10"},
    "gpt-4o-realtime-preview-2024-10-01": {"shutoff": "2025-10-10"},
    "o1-preview-2024-09-12": {"shutoff": "2025-07-28", "alias": "o1-preview"},
    "gpt-4.5-preview": {"shutoff": "2025-07-14"},
    "gpt-4-32k-0613": {"shutoff": "2025-06-06", "alias": "gpt-4-32k"},
    "gpt-4-32k-0314": {"shutoff": "2025-06-06"},
    "gpt-4-1106-vision-preview": {"shutoff": "2024-12-06", "alias": "gpt-4-vision-preview"},
    "gpt-3.5-turbo-0613": {"shutoff": "2024-09-13", "ft_shutoff": "2026-10-23"},
    "gpt-3.5-turbo-16k-0613": {"shutoff": "2024-09-13", "ft_shutoff": "2026-10-23", "alias": "gpt-3.5-turbo-16k"},
    "gpt-3.5-turbo-0301": {"shutoff": "2024-09-13", "ft_shutoff": "2026-10-23"},
    # GPT-3 all dead
}

fields of dict/object:
“shutoff” - turnoff date on API
“ft_shutoff” - when fine-tuning versions of a model will be shut off
“alias” - the pointer to the model being shut off with no new destination

16

In the previous reply, I revised to give a high-quality assurance pass to make sure every alias/pointer that is going to die is also useful for shutoff model gating, along with fine tuning models by the model’s end date.

After all other models are hit with shutoff - including as recent as gpt-4.1-nano fine-tunings (you can still train) shutoff in three months - I thought it would be interesting to see what still remains to chat with.

PERSISTED_MODELS: dict[str, dict[str, str | None]] = {
    # GPT-5.6 Family
    "gpt-5.6-sol":                          {"alias": "gpt-5.6"},
    "gpt-5.6-terra":                        {"alias": None},
    "gpt-5.6-luna":                         {"alias": None},
    # GPT-5.5 Family
    "gpt-5.5-2026-04-23":                   {"alias": "gpt-5.5"},
    "gpt-5.5-pro-2026-04-23":               {"alias": "gpt-5.5-pro"},
    "chat-latest":                          {"alias": None},
    # GPT-5.4 Family
    "gpt-5.4-2026-03-05":                   {"alias": "gpt-5.4"},
    "gpt-5.4-mini-2026-03-17":              {"alias": "gpt-5.4-mini"},
    "gpt-5.4-nano-2026-03-17":              {"alias": "gpt-5.4-nano"},
    "gpt-5.4-pro-2026-03-05":               {"alias": "gpt-5.4-pro"},
    # GPT-5.3 Codex
    "gpt-5.3-codex":                        {"alias": None},
    # GPT-5.2 Family
    "gpt-5.2-2025-12-11":                   {"alias": "gpt-5.2"},
    "gpt-5.2-pro-2025-12-11":               {"alias": "gpt-5.2-pro"},
    # GPT-5.1 (no pro)
    "gpt-5.1-2025-11-13":                   {"alias": "gpt-5.1"},
    # GPT-4.1 Family
    "gpt-4.1-2025-04-14":                   {"alias": "gpt-4.1"},  # with fine-tuning
    "gpt-4.1-mini-2025-04-14":              {"alias": "gpt-4.1-mini"},  # with fine-tuning
    # GPT-4o Family
    "gpt-4o-2024-11-20":                    {"alias": None},
    "gpt-4o-2024-08-06":                    {"alias": "gpt-4o"},  # with fine-tuning
    "gpt-4o-mini-2024-07-18":               {"alias": "gpt-4o-mini"},  # with fine-tuning
    # GPT-5 Search (chat completions web search)
    "gpt-5-search-api-2025-10-14":          {"alias": "gpt-5-search-api"},
}

OpenAI announced a bunch with three month notice, so this might not be a picture of the end of the year. They’ll probably kill 5.3-codex next with short notice.

Documentation: I also note that gpt-4o-search-preview-2025-03-11 to be shut off this month, for web search on chat completions where there are no internal tools, has an ineffective replacement in the documentation. It should be gpt-5-search-api - for however long that lasts - not gpt-5.4-mini