AI model retirements: what shuts down Nov 2026 to Feb 2027

As of October 24, 2026, 06:20 JST (October 23, 21:20 UTC). Every row is from the vendor's own deprecation, lifecycle, or changelog page, read between 05:36 and 05:46 JST on October 24; the OpenAI, Anthropic, GitHub, Microsoft, and AWS pages were read again at 05:56 with no change, and Google's at 06:20. Each source below carries its own date and the minutes it was read. A date that passes, and a date a vendor adds or moves, gets a dated line under Updates to this page.

On the vendors' own services, the next three shutdowns are: GitHub Copilot drops Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra on November 16, 2026; Google's earliest date for five Gemini API preview audio models is November 17; and Anthropic retires claude-sonnet-4-5-20250929 on November 30.

Their replacements, as each vendor names them: Claude Opus 5.5, GPT-6.1 Sol, or GPT-6 Luna in Copilot; gemini-3.8-live or gemini-3.8-flash-tts in the Gemini API; claude-sonnet-5-5 in the Claude API. Further out, on January 14, 2027, OpenAI removes GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini from the API and names gpt-6-luna as their replacement. GPT-4o already left the ChatGPT app on February 13, 2026; that was the app, not the API.

The same model can stop on another date on another cloud. Amazon Bedrock drops five Claude 3.x models in some Regions on November 13, 2026, and keeps Claude Sonnet 4.5 until April 8, 2027. Azure retires o3 and o4-mini on November 19, 2026, three weeks before OpenAI's own date for o3.

Below: one table of every date from October 24, 2026 to February 28, 2027 for OpenAI, Anthropic, and Google models, on their own services, in GitHub Copilot, on Azure, and on Bedrock, with the other publishers' dates on Azure and Bedrock after it; then what shut down in the last few days, each vendor's rules, the cloud dates that differ, the apps versus the APIs, and how to find the ids your code calls.

The calendar, November 2026 to February 2027

Sorted by date. "Surface" says where the date applies: a vendor's API, an app, or another cloud's copy of the model. The replacement is the one the vendor names on the same page; "none listed" means the page names none. The second column links each vendor's page once and gives the date of the notice or of the page itself; the full list is under Sources.

Other publishers' models on the same two clouds, in the same window. Microsoft's schedule: Meta's Llama-4-Maverick-17B-128E-Instruct-FP8 (November 19, 2026), DeepSeek-V4-Flash-0731, Preview (December 3), Mistral's Ministral-3B and mistral-medium-2505 (December 9), Cohere-parse-v5, Preview (December 15), Microsoft's five Phi-4 models (January 17, 2027), Kimi-K2.5, Preview (January 26), Meta's Llama-3.3-70B-Instruct (January 31), and cohere-command-a, Cohere-embed-v3-english, Cohere-embed-v3-multilingual, and mistral-small-2503 (February 15). Bedrock's Legacy page: AI21's Jamba 1.5 Large and Mini (November 26, 2026) and TwelveLabs' Marengo Embed v2.7 (November 30, 2026, in us-east-1, eu-west-1, and ap-northeast-2). Fireworks models on Foundry have their own catalog page, not read for this page. The next dates after February: Bedrock moves a long list of Meta, Mistral, Google Gemma, DeepSeek, Qwen, and other models to end of life on March 30, 2027; Google's earliest date for gemini-2.5-flash-image is March 15, 2027; and OpenAI removes gpt-5.1, gpt-5.3-codex, and gpt-5.4-nano from the API on April 1, 2027.

Shut down in the last few days

If something broke this week, check these first.

• OpenAI API, October 23, 2026. gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4-1106-preview, gpt-4.1-nano, gpt-4o-2024-05-13, gpt-image-1, o1, o1-pro, o3-mini, o4-mini, and fine-tunes of gpt-3.5-turbo, gpt-4, gpt-4.1-nano, o4-mini, babbage-002, and davinci-002. OpenAI's substitutes are gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna by model, and gpt-image-2.5-sunburst or gpt-image-2.5-flare for images. When this page was read, it was still October 23 in UTC, and OpenAI's page still listed these under "Upcoming deprecations".
• Gemini API, October 22 and 23, 2026. veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview, and gemini-omni-flash-preview on October 22, all four replaced by gemini-omni-1.1-flash. Google's page shades these rows gray, which it defines as "Already-shutdown models". The deep-research-pro-preview-12-2025 agent's date was October 23 (replacement deep-research-preview-04-2026); its row wasn't gray yet when read at 06:20 JST.
• Azure OpenAI, October 23, 2026. gpt-image-1 2025-04-15, a Preview, per Microsoft's schedule. No replacement listed.
• GitHub Copilot, October 19, 2026. Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5, "across most GitHub Copilot experiences". Suggested: Gemini 3.8 Flash, GPT-6.1 Sol, GPT-6 Luna, and Grok 4.7.
• GitHub Copilot, October 2, 2026. Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7. Suggested: Gemini 3.8 Flash, Kimi K3, and Claude Opus 5.5.

Anthropic retired no Claude API model in October; its last was claude-opus-4-1-20250805, on August 5, 2026.

OpenAI API

OpenAI's page holds most of the dates in the window, and the October 14 notice is the one that covers GPT-4o and GPT-4.1: "On October 14, 2026, we notified developers using certain GPT-4 series models of their deprecation and removal from the API on January 14, 2027." It adds: "Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models."

The short names follow the snapshots. The deprecations page lists dated snapshots such as gpt-4o-2024-08-06, not the short names most code sends. Each short name's model page names its default snapshot (checked 05:40 JST), and every one of these is on the list:

The page doesn't say whether a short name stops answering on the date or moves to another snapshot. Plan as if it stops.

The notice was shorter than OpenAI's usual minimum. The same page promises "Generally available models: At least 6 months" of notice, "Unless safety or compliance concerns require a faster timeline". The GPT-4 series notice of October 14 gives three months to January 14, and the page gives no reason. The other model notices with dates in the window gave about six months, except the text-to-speech notice of October 1, which says "with at least three months' notice". The page also says "In some cases, developers may be able to provision dedicated capacity for continued access after a model's shutdown date", through OpenAI's sales team.

Not only models. Evals turns read-only on October 31 and shuts down with Agent Builder and reusable prompts on November 30. On January 6, 2027, existing customers "will no longer be able to create new fine-tuning jobs"; inference on fine-tuned models continues until their base model's own date.

Checking from code. OpenAI's page says the List models endpoint (GET /v1/models) returns an optional shutdown_date field for each model, "as a YYYY-MM-DD string, or null if no shutdown date has been announced".

Anthropic Claude API

One Claude API model has a retirement date in the window: claude-sonnet-4-5-20250929, deprecated on September 30, 2026 and retiring on November 30, 2026, with claude-sonnet-5-5 as the replacement. Anthropic's page says "Requests to models past the retirement date will fail."

Four more models show a "Tentative retirement date" in the window or just before it, each worded "Not sooner than": claude-haiku-4-5-20251001 (October 15, 2026, already passed), claude-opus-4-5-20251101 (November 24, 2026), claude-opus-4-6 (February 5, 2027), and claude-sonnet-4-6 (February 17, 2027). All four are marked Active, and none has a deprecation notice on the page. Anthropic promises "at least 60 days' notice before model retirement for publicly released models", so a notice sent on October 24 couldn't retire any of them before December 23, 2026. That date is this page's arithmetic, not Anthropic's.

Where these dates apply. "The dates on this page apply to Anthropic-operated platforms: the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Microsoft Foundry." Amazon Bedrock and Claude Models on Google Cloud (Google managed) "set their own retirement schedules". Bedrock's are below; Google Cloud's own page was not read for this page.

Checking which models you call. In the Claude Console, open Usage and click Export; Anthropic says the CSV breaks usage down "by API key and model".

One change in the window isn't a model: in API version 2026-10-21, sending temperature, top_p, or top_k "Returns a 400 error", per the same page's parameter table.

Google Gemini API

Every Gemini date is a floor, in Google's words: "The shutdown dates listed in the table indicate the earliest possible dates on which a model might be retired", and "We will communicate the exact shutdown date to users with advance notice". The page was last updated on October 22, 2026.

In the window, five preview models share the earliest date of November 17, 2026: three text-to-speech models move to gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts, and two Live API audio models move to gemini-3.8-live (calendar). The first model without "preview" in its name to carry a date is gemini-2.5-flash-image, on March 15, 2027, with gemini-3.1-flash-lite-image as its replacement.

Gemini 2.5 Pro and Flash have no shutdown date, but new users can't start on them. gemini-2.5-pro, gemini-2.5-flash, and gemini-2.5-flash-lite show "No shutdown date announced", and Google's note on them reads: "To ensure reliable performance for everyone, we are limiting access to the 2.5 models to users who have actively used them in the past. These models are not deprecated and will continue to be served until further notice through the API. For any new projects, use our latest models: 3.5 Flash-Lite or 3.8 Flash." So existing 2.5 code has no deadline yet, and a new project starts on 3.5 Flash-Lite or 3.8 Flash.

Two names now answer with another model. The rows for gemini-3.5-flash and gemini-3.7-flash say the endpoint "is not shutting down" and "automatically routes API requests to" gemini-3.6-flash and gemini-3.8-flash respectively. Your calls keep working, but the model behind them has changed.

The page covers the Gemini API. Vertex AI keeps its own model lifecycle pages, which were not read for this page.

GitHub Copilot

GitHub's changelog of October 15, 2026: "We will deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on November 16th, 2026". The three are Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. In Copilot, the deprecation date is the day a model leaves: GitHub's docs list the October 2 models and five of the six October 19 models (not Grok 4.5) under "Model retirement history", with those same dates as their "Retirement date". The docs didn't list the November 16 models yet when read at 05:38 JST.

• The exception: "Claude Opus 4.8 will remain available to individual GitHub Copilot subscribers on annual plans so they retain an Opus offering."
• Business and Enterprise: "Under default model enablement, the suggested alternatives are automatically enabled for Copilot Enterprise and Copilot Business customers unless an administrator has turned off the global default or explicitly disabled the model." If your organization turned the default off, enable the alternatives in the model policies in Copilot settings.
• Nothing to remove: "No action is required to remove the models after they have been deprecated."

Copilot's dates are Copilot's own. Claude Opus 4.8 leaves Copilot on November 16, while Anthropic's API lists claude-opus-4-8 as Active, not to retire sooner than May 28, 2027. Code that calls Anthropic's API directly isn't affected by the Copilot date.

Same model, other cloud: Azure and Bedrock dates

Running a vendor's model through Microsoft Foundry (Azure OpenAI) or Amazon Bedrock puts you on that cloud's date, which can be earlier or later. Bedrock says so plainly: "For Amazon Bedrock usage, only the dates on this page apply."

"Not listed" means the model isn't on that cloud's retirement or lifecycle pages. Bedrock's lifecycle page says one OpenAI model there, GPT-6.1 Sol, "follows the OpenAI first-party model's lifecycle".

Azure: what happens on the date. For Standard, Global Standard, and Data Zone Standard deployments, Microsoft upgrades the deployment to the replacement its schedule lists, "where supported", "on a rolling, region-by-region basis". Microsoft names that replacement "approximately 90–120 days before retirement", so several rows in the window, the GPT-5 models of February 9 among them, list none yet. "Provisioned deployments are NOT auto-upgraded." A deployment set to NoAutoUpgrade "stops working at retirement", and a retired model's "inference returns 410 Gone". Microsoft also says: "Retirement dates aren't extendable."

Bedrock: what happens before the date. In the Legacy period, "New customers can't use Legacy models and existing customers may lose access to Legacy models after 15 days of inactivity." In the public extended access part of it, "you should expect higher pricing, which will be set by the model provider." At end of life, "Migration will not happen automatically."

Where Microsoft's and Anthropic's dates disagree

Anthropic says its dates apply to Microsoft Foundry. Microsoft's schedule, last updated September 23, 2026, gives firm retirement dates for four Claude models that Anthropic lists as Active, three of them on other days:

For Opus 4.5 the day is the same, but Anthropic's is a floor and Microsoft's is a retirement date; under Anthropic's 60 days' notice, its own date can't land on November 24 without a notice. Neither page says which one wins. If you run Claude on Foundry, plan for Microsoft's date, or the earlier one where they differ, and ask your Microsoft account team. Microsoft's schedule also lists claude-opus-5-5-fast, version 2, which Microsoft hosts on Azure, with a retirement date of January 15, 2027; that model isn't on Anthropic's page.

The apps are not the API

"When does GPT-4o retire?" has two answers, and OpenAI's own pages give both.

ChatGPT, the app. OpenAI's help article "Retiring GPT-4o and other ChatGPT models" opens: "GPT-4o and additional models were deprecated in ChatGPT on February 13, 2026. These models will continue to be available in the API." That day, ChatGPT retired GPT-4o, GPT-4.1, GPT-4.1 mini, OpenAI o4-mini, and GPT-5 (Instant and Thinking); GPT-5.1 Instant, Thinking, and Pro followed on March 11, 2026. Business, Enterprise, and Edu workspaces kept GPT-4o in Custom GPTs until April 3, 2026. The article says ChatGPT Voice and ChatGPT Images were "not changing as part of this update".

The API. The help article, marked "Updated: 3 days ago" (about October 21) when read at 05:44 JST on October 24, still opens with "These models will continue to be available in the API", and further down says "we'll provide advance notice ahead of any future API retirements." OpenAI's API deprecations page now has such a notice for four of them: on October 14, 2026, OpenAI announced that GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini leave the API on January 14, 2027. For code that calls the API, the API deprecations page governs.

Claude apps. Anthropic's deprecations page "lists all API deprecations". It gives no dates for models in the Claude apps, and no Anthropic page found for this article does.

Gemini app. Google's deprecations page covers the Gemini API and says nothing about the Gemini app.

Copilot is itself an app with its own dates, as above.

Find the model ids your code calls

1. Search your code and config for the retiring ids

From your repository root, this lists OpenAI ids with a date in the window, including the short names:

grep -rnE "(^|[^a-z0-9.-])(gpt-4o|gpt-4o-mini|gpt-4\.1|gpt-4\.1-mini|gpt-5|gpt-5-mini|gpt-5-nano|gpt-5-pro|o3|o3-pro|tts-1|tts-1-hd|whisper-1|gpt-image-1\.5|gpt-image-1-mini|chatgpt-image-latest|gpt-4o-transcribe|gpt-4o-transcribe-diarize|gpt-4o-mini-transcribe|gpt-4o-mini-tts|gpt-realtime|gpt-realtime-mini|gpt-4o-realtime|gpt-4o-mini-realtime|gpt-audio|gpt-audio-mini|gpt-4o-audio|gpt-4o-mini-audio)(-preview)?(-20[0-9]{2}-[0-9]{2}-[0-9]{2})?([^a-z0-9.-]|$)" . --exclude-dir=node_modules --exclude-dir=.git

And this one the Claude, Gemini, and Bedrock ids:

grep -rnE "claude-sonnet-4-5|claude-opus-4-[156]|claude-opus-5-5-fast|claude-haiku-4-5|claude-3-|claude-sonnet-4-6|gemini-3\.1-flash-(tts|live)-preview|gemini-2\.5-(flash|pro)-preview-tts|gemini-2\.5-flash-native-audio-preview-12-2025" . --exclude-dir=node_modules --exclude-dir=.git

The first pattern skips gpt-5.6-sol, gpt-realtime-2.1, gpt-audio-1.5, and other newer names on purpose. A bare o3 can also match unrelated text, so read each hit.

2. Check where a model id lives outside your code

Model ids also sit in .env files, CI and hosting variables, and proxy configs such as a LiteLLM model_list. On Azure OpenAI, your code calls a deployment name you chose, so open each deployment and read its model version. In Claude Code, the model comes from /model, claude --model, the ANTHROPIC_MODEL environment variable, or the model field in settings.json. A full id pinned in ANTHROPIC_DEFAULT_SONNET_MODEL, ANTHROPIC_DEFAULT_OPUS_MODEL, or ANTHROPIC_DEFAULT_HAIKU_MODEL keeps that model until you change it.

3. Ask the vendor what you actually call

OpenAI's GET /v1/models returns shutdown_date for each model. Anthropic's Console exports usage by API key and model under Usage, then Export. Azure's Models API returns lifecycleStatus and deprecation.inference for each model in a Region. Bedrock's ListFoundationModels and GetFoundationModel return modelLifecycle, and Bedrock says to call them "in that Region", because the state can differ by Region.

4. Switch, test, and deploy before the date

Point each call at the vendor's replacement from the calendar, then run your own prompts against it before you ship. Anthropic's page advises "thorough testing of your applications with the new models well before the retirement date", and Microsoft's says to compare "quality, latency, and cost together rather than relying on public benchmarks alone". If you test with OpenAI Evals, move the tests first: Evals turns read-only on October 31 and shuts down on November 30.

What to switch to

Each vendor's own recommended replacement, by model family.

A vendor's replacement is a starting point. Microsoft's page notes the replacement "doesn't limit your manual migration choices", and the same holds for the others: any current model that passes your tests will do.

Updates to this page

• October 24, 2026, 06:20 JST: first version. Vendor pages read between 05:36 and 05:46 JST, five of them again at 05:56, and Google's at 06:20.

Each vendor page is read again at least once a week, and on November 17 and December 1, the days after the Copilot change and the Claude Sonnet 4.5 retirement. A notice found on any of those reads gets a dated line here the same day. The rows most likely to change: Anthropic's four "not sooner than" dates, which become firm only with a notice; the Gemini preview dates, which Google calls earliest dates; GitHub's docs, which don't list the November 16 Copilot models yet; and Microsoft's schedule, whose Claude dates differ from Anthropic's.

Other dates the same months: End of life in November 2026: .NET 8 and 9, PostgreSQL 14 and Node.js 26 LTS on October 28: which version to run?.

Frequently asked questions

When does the GPT-4o API shut down?

On January 14, 2027. OpenAI's notice of October 14, 2026 removes gpt-4o-2024-08-06 and gpt-4o-2024-11-20 from the API on January 14, 2027, and OpenAI's model page names gpt-4o-2024-08-06 as the default snapshot behind gpt-4o. The older gpt-4o-2024-05-13 was scheduled to shut down on October 23, 2026. On Azure OpenAI, GPT-4o 2024-08-06 and 2024-11-20 retire on April 14, 2027.

Is GPT-4o still available in ChatGPT?

No. OpenAI's help article says GPT-4o "and additional models were deprecated in ChatGPT on February 13, 2026", and Custom GPTs in Business, Enterprise, and Edu workspaces kept it until April 3, 2026: "After April 3, GPT-4o will be fully retired across all plans." The API date is separate: January 14, 2027.

What replaces GPT-4o and GPT-4.1 in the API?

gpt-6-luna, for GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini alike, per OpenAI's deprecations page.

When is Claude Sonnet 4.5 retired?

On November 30, 2026, on the Claude API and the other Anthropic-operated platforms; the replacement is claude-sonnet-5-5. On Amazon Bedrock, anthropic.claude-sonnet-4-5-20250929-v1:0 reaches end of life on April 8, 2027, with public extended access from January 8, 2027.

Is Claude Opus 4.5 being retired on November 24, 2026?

On the Claude API, not as of October 24, 2026. Anthropic lists claude-opus-4-5-20251101 as Active with a tentative retirement date "Not sooner than November 24, 2026" and no deprecation notice. With Anthropic's 60 days' notice, a notice sent on October 24 couldn't retire it before December 23, 2026. On Microsoft Foundry, yes: Microsoft's schedule lists claude-opus-4-5, version 1, for retirement on November 24, 2026.

Which Copilot models are deprecated on November 16, 2026?

Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra, across all GitHub Copilot experiences. GitHub suggests Claude Opus 5.5 for both Claude models and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. Individual subscribers on annual plans keep Claude Opus 4.8.

Are Gemini 2.5 Pro and Gemini 2.5 Flash being shut down?

No date is set, but access is limited. Google's page shows "No shutdown date announced" for gemini-2.5-pro, gemini-2.5-flash, and gemini-2.5-flash-lite, and says: "we are limiting access to the 2.5 models to users who have actively used them in the past. These models are not deprecated and will continue to be served until further notice through the API. For any new projects, use our latest models: 3.5 Flash-Lite or 3.8 Flash."

What happens when my code calls a retired model?

The call fails. OpenAI: at the shutdown, "the model or endpoint will no longer be accessible." Anthropic: "Requests to models past the retirement date will fail." Azure: "inference returns 410 Gone", unless a Standard deployment was auto-upgraded first. Bedrock: "requests made to this version will fail."

Do fine-tuned models retire too?

On OpenAI, yes, on the base model's date: "Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models." On Azure, fine-tuned models retire in two phases, training and then deployment; deployments of fine-tuned GPT-4o 2024-08-06 run until October 1, 2027.

How can I check a model's shutdown date from code?

On OpenAI, call GET /v1/models and read each model's shutdown_date. On Azure, call the Models API for your Region and read lifecycleStatus and deprecation.inference. On Bedrock, call ListFoundationModels or GetFoundationModel in your Region and read modelLifecycle. Anthropic and Google publish their dates on the pages linked below.

Sources

Read on October 24, 2026, times in JST (UTC+9).

• OpenAI: Deprecations, no page date; notices from April 22, 2026 to October 14, 2026 (05:36, 05:38, 05:56)
• OpenAI model pages: GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini, GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 pro, o3, o3-pro, and gpt-6-luna (05:40)
• OpenAI Help Center: Retiring GPT-4o and other ChatGPT models, marked "Updated: 3 days ago", about October 21 (05:44)
• Anthropic: Model deprecations, no page date; notices to September 30, 2026 (05:36, 05:38, 05:56)
• Google: Gemini deprecations, last updated October 22, 2026 (05:36; again 05:43–05:46 and 06:20)
• GitHub changelog: Upcoming deprecation of selected GitHub Copilot models in mid-November, October 15, 2026 (05:38), and the Copilot changelog list, with no newer retirement through October 22 (05:56)
• GitHub changelog: GitHub Copilot models deprecated on October 19, 2026, October 19, 2026 (05:38)
• GitHub changelog: Selected models in GitHub Copilot deprecated, October 2, 2026 (05:38)
• GitHub Docs: Supported AI models in GitHub Copilot, "Model retirement history" (05:38)
• Microsoft Foundry: Model retirement schedule, last updated September 23, 2026 (05:41, 05:56)
• Microsoft Foundry: Models lifecycle and support policy, last updated July 24, 2026 (05:39)
• Amazon Bedrock: Model lifecycle (Legacy), no page date (05:41, 05:56) and Model lifecycle (05:39)
• Claude Code docs: Model configuration, no page date (05:45)