The two biggest model vendors publish retirement dates, and they do not publish them anywhere near each other.
This page puts both tables in one place, taken from the vendors' own deprecation documentation rather than from
news coverage, and is updated when either page changes.
WHAT TO KNOW FIRST
● OpenAI notice period: at least 6 months for generally available models, 3 months for specialised variants, as little as 2 weeks for preview models
● Anthropic notice period: at least 60 days for publicly released models
● Next cliff: 30 November 2026 - Claude Sonnet 4.5 retires and three OpenAI tools shut down the same day
● Already breaking: temperature, top_p and top_k return a 400 error on Anthropic API version 2026-10-21
● Biggest December date: 11 December 2026 - the entire GPT-5 family and o3 shut down
The notice gap is the thing to plan around
OpenAI commits to at least six months of notice before retiring a generally available model. Anthropic commits to
at least sixty days. That is a threefold difference in migration runway, and it is the single most useful fact on
this page if you are choosing where to build.
It cuts both ways. Anthropic publishes a tentative retirement date for every active model, often
eighteen months out, so you can see the whole fleet's expected lifespan at a glance. OpenAI does not publish
forward dates for models that have not been deprecated yet - you find out when the notice lands. Longer notice,
less visibility.
OpenAI shutdown schedule
| Date | Model or endpoint | Replacement | Status |
|---|
| 31 Oct 2026 | Evals (existing) | Become read-only | Past |
| 30 Nov 2026 | v1/prompts API, reusable prompt objects | Move prompt content into application code | Due |
| 30 Nov 2026 | Evals dashboard and API | Promptfoo migration path linked | Due |
| 30 Nov 2026 | Agent Builder | Agents SDK or ChatGPT Workspace Agents | Due |
| 1 Dec 2026 | gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest | gpt-image-2.5-sunburst or -flare | Due |
| 11 Dec 2026 | gpt-5, gpt-5-2025-08-07 | gpt-5.6-sol | Due |
| 11 Dec 2026 | gpt-5-mini | gpt-5.6-terra | Due |
| 11 Dec 2026 | gpt-5-nano | gpt-5.6-luna | Due |
| 11 Dec 2026 | gpt-5-pro | gpt-5.6-sol (reasoning.mode: pro) | Due |
| 11 Dec 2026 | o3, o3-pro | gpt-5.6-sol | Due |
| 6 Jan 2027 | tts-1, tts-1-hd, gpt-4o-mini-tts | gpt-realtime-2.1-mini | Scheduled |
| 6 Jan 2027 | Self-serve fine-tuning job creation | Not stated | Scheduled |
| 14 Jan 2027 | gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini | gpt-6-luna | Scheduled |
| 14 Jan 2027 | gpt-4.1, gpt-4.1-mini | gpt-6-luna | Scheduled |
| 20 Jan 2027 | gpt-realtime, gpt-audio and 4o variants | gpt-realtime-2.1 or gpt-audio-1.5 | Scheduled |
| 26 Feb 2027 | whisper-1, gpt-4o-transcribe family | gpt-live-transcribe or gpt-transcribe | Scheduled |
| 26 Feb 2027 | /v1/audio/translations | v1/realtime/translations | Scheduled |
| 1 Apr 2027 | gpt-5.1, gpt-5.3-codex | gpt-6-sol | Scheduled |
| 1 Apr 2027 | gpt-5.4-nano | gpt-6-luna | Scheduled |
From OpenAI's own deprecations page, read 26 October 2026. Entries before 26 October are omitted except
where they are still being discovered in production - the 23 October 2026 batch retired gpt-3.5-turbo, gpt-4,
gpt-4-turbo, o1, o1-pro, o3-mini, o4-mini, gpt-image-1 and their fine-tunes.
Anthropic model status
| Retirement date | API model string | Replacement | State |
|---|
| 30 Nov 2026 | claude-sonnet-4-5-20250929 | claude-sonnet-5-5 | Deprecated |
| Not before 24 Nov 2026 | claude-opus-4-5-20251101 | No replacement named | Active |
| Not before 15 Oct 2026 | claude-haiku-4-5-20251001 | No replacement named | Active |
| To be announced | claude-mythos-preview | Not stated | Deprecated |
| Not before 5 Feb 2027 | claude-opus-4-6 | No replacement named | Active |
| Not before 17 Feb 2027 | claude-sonnet-4-6 | No replacement named | Active |
| Not before 16 Apr 2027 | claude-opus-4-7 | No replacement named | Active |
| Not before 28 May 2027 | claude-opus-4-8 | No replacement named | Active |
| Not before 9 Jun 2027 | claude-fable-5, claude-mythos-5 | No replacement named | Active |
| Not before 30 Jun 2027 | claude-sonnet-5 | No replacement named | Active |
| Not before 24 Jul 2027 | claude-opus-5 | No replacement named | Active |
| Not before 1 Sep 2027 | claude-fable-5-1, claude-mythos-5-1 | No replacement named | Active |
| Not before 22 Sep 2027 | claude-opus-5-5 | No replacement named | Active |
| Not before 28 Sep 2027 | claude-sonnet-5-5 | No replacement named | Active |
| Not before 7 Oct 2027 | claude-haiku-5-5 | No replacement named | Active |
From Anthropic's own model deprecations page, read 26 October 2026. "Not before" dates are tentative and
apply to Anthropic-operated platforms only - Amazon Bedrock and Google-managed Claude models set their own
schedules and can differ.
The 30 November cliff
Four things stop working on the same day. Claude Sonnet 4.5
(claude-sonnet-4-5-20250929) retires from the Claude API, replaced by
claude-sonnet-5-5. It was deprecated on 30 September, giving exactly
the sixty days Anthropic commits to.
On the OpenAI side, Agent Builder, the Evals dashboard and API, and the
v1/prompts API all shut down. The Evals API is the one that bites quietly: evaluation suites run
in CI, so the failure surfaces as a red pipeline rather than a broken product, often weeks later. OpenAI links a
Promptfoo migration path for it. For v1/prompts there is no replacement service - the guidance is to move prompt
content into your application code.
The parameter change already returning errors
Separate from model retirement, and easy to miss: Anthropic removed
temperature, top_p and
top_k in API version 2026-10-21. On that version they
return a 400 error. On earlier versions they still error when set to a non-default value on
Claude 4.7 and later models. The recommended approach is to omit them and steer behaviour through prompting
instead.
How to audit what you are actually calling
Anthropic provides a usage export that breaks calls down by API key and model: Claude Console, Usage, Export, then
read the CSV. That is the fastest way to find a deprecated model string buried in a config file nobody has opened
since it was written - which is where these usually hide.
Frequently asked questions
When does Claude Sonnet 4.5 stop working?
30 November 2026 on the Claude API. The model string is claude-sonnet-4-5-20250929 and the recommended replacement is claude-sonnet-5-5. Amazon Bedrock and Google-managed deployments set their own dates.
When is GPT-5 being shut down?
11 December 2026, along with gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro. Replacements are gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna depending on tier.
How much notice does OpenAI give before retiring a model?
At least six months for generally available models, at least three months for specialised variants such as chat, Codex and deep research, and as little as two weeks for preview models. Safety or compliance issues can force faster retirement.
How much notice does Anthropic give?
At least sixty days for publicly released models, with email notification to customers running active deployments. Anthropic also publishes a tentative retirement date for every active model.
Why am I getting a 400 error on temperature?
Anthropic removed temperature, top_p and top_k in API version 2026-10-21. They now return a 400 error on that version. Omit them and guide behaviour through prompting.
What happens if I call a retired model?
The request fails. Anthropic states plainly that requests to retired models fail; OpenAI shuts the endpoint down. Neither silently falls back to a newer model.
When is GPT-4o being retired?
14 January 2027 for gpt-4o-2024-08-06, gpt-4o-2024-11-20 and gpt-4o-mini-2024-07-18, replaced by gpt-6-luna. The older gpt-4o-2024-05-13 snapshot went on 23 October 2026.
Sources
OpenAI dates and notice policy from OpenAI Deprecations. Anthropic dates, lifecycle states and the parameter removal from Anthropic Model deprecations. Both read 26 October 2026. See also our AI API pricing tracker.
Last updated 26 October 2026. Dates and prices change - verify against the vendor before acting.