What is the difference between deprecation and retirement?

Deprecation (or, on Amazon Bedrock, the Legacy state) means a model is still callable but is scheduled to go away: the vendor has flagged it, usually named a recommended replacement, and set a date. Retirement - called shutdown by OpenAI and the Gemini API, retired by Anthropic and Vertex AI, and EOL by Amazon Bedrock - is the date the model stops working. Before that date, existing code keeps running; after it, calls fail.

How do vendors announce model deprecations?

The notice periods differ. Anthropic provides at least 60 days notice before retirement for publicly released models. OpenAI gives at least 6 months for generally available models, at least 3 months for specialized variants, and as little as about 2 weeks for preview models. The Gemini API gives at least 2 weeks for preview and latest aliases and treats its table dates as the earliest possible shutdown. For models launched before 2026-09-07, Amazon Bedrock keeps a model at least 12 months on the platform and at least 6 months in the Legacy state before EOL; for models launched on or after that date, model cards state, where applicable, an EOL no sooner than date and a Legacy period of 6 months or 45 days. Vendors notify affected customers by email and update their documentation; larger changes are also announced on blogs.

What happens to API calls to a retired model?

They fail. Anthropic states that requests to retired models will fail. Amazon Bedrock states that after EOL requests made to this version will fail. The Gemini API states that a shut-down model is completely turned off, and the endpoint is no longer available. OpenAI states that a shut-down model or endpoint will no longer be accessible. Migration does not happen automatically - you must update your application code to a supported model before the date.

How should I plan a model migration?

Find your model's row for the specific platform you call, because the same model can have different dates on different platforms. Note the retirement, shutdown, or EOL date and the vendor's recommended replacement, test the replacement, and switch before the date. Step-by-step migration mechanics for Claude models are covered in the Anthropic Claude Model Migration Guide, and the LLM Token Counter and Context Budget Planner can help estimate context and token budgets for a candidate replacement.

AI Model Deprecation and Lifecycle Calendar - Anthropic, OpenAI, Google, and Amazon Bedrock

First Published: 2026-07-05
Last Updated: 2026-10-10

This page is a single, cross-vendor calendar of announced deprecation, retirement, shutdown, and end-of-life (EOL) dates for AI models on the four platforms most teams run in production: the Anthropic API, OpenAI, Google (the Gemini API and Vertex AI), and Amazon Bedrock. Model end-of-life is a recurring operational event for every AI team, yet each vendor publishes its schedule on a different page and defines the terms differently. The goal here is to keep those official facts in one place and to translate each vendor's vocabulary into a common frame.

The First Published and Last Updated dates shown above indicate the currency of this page. Companion references on hidekazu-konishi.com:

Overview

This page collects, in one view, the model lifecycle information that each vendor publishes on its own official documentation:

Anthropic — the Model deprecations page and the Models overview for the Anthropic API (also offered as Claude Platform on AWS and Microsoft Foundry).
OpenAI — the Deprecations page and the Models page.
Google — the Gemini API deprecations page (Google AI for Developers) and the Vertex AI model-versioning and deprecations pages (Google Cloud).
Amazon Bedrock — the Model lifecycle page in the Amazon Bedrock User Guide.

This page is reviewed and updated quarterly, and whenever major deprecations are announced. Because model lifecycle dates change — vendors add rows, adjust dates, and retire models — always confirm the current dates on the official pages linked in each section before you act.

The scope of this page is limited to the four platforms above and to dates that are officially published. I do not include unannounced or speculated dates, pricing, or performance comparisons. Migration mechanics (how to move code from one model to another) are covered in the dedicated Anthropic Claude Model Migration Guide; this page is about when a model changes state and what the vendor lists as its replacement.

How to Read This Page

Lifecycle terminology differs by vendor

Every vendor uses slightly different words for the same two ideas: "still usable but scheduled to go away" and "no longer usable." The table below maps each vendor's official terms onto that common frame. The definitions are quoted from each vendor's official documentation.

Concept	Anthropic	OpenAI	Google Gemini API	Google Vertex AI	Amazon Bedrock
Still callable, scheduled to end	Deprecated ("still functional but no longer recommended"; a recommended replacement and retirement date are assigned). Also Legacy ("will no longer receive updates and may be deprecated in the future").	Deprecation ("the process of retiring a model or endpoint"; becomes deprecated immediately when announced). Legacy models "no longer receive updates."	No separate "deprecated" state; a model is scheduled with a shutdown date. Stability labels: stable, preview, experimental, latest.	Latest stable vs legacy stable (an older, still-supported stable version).	Legacy ("in the Legacy state for at least 6 months before the EOL date"; new customers cannot use it). For models launched on or after 2026-09-07, the Legacy period is 6 months or 45 days, as shown on the model card where applicable (GPT-6.1 Sol follows OpenAI's first-party lifecycle terms instead).
No longer callable	Retired ("no longer available for use. Requests to retired models will fail").	Shutdown / Sunset ("completely turned off"; "will no longer be accessible").	Shutdown ("completely turned off, and the endpoint is no longer available").	Retired ("past its retirement date and has been permanently deactivated").	EOL (End-of-Life) ("no longer available for use in all AWS Regions and requests made to this version will fail").

The meaning of the dates

An announcement / deprecation date is when the vendor first flags a model. The model is usually still fully callable at this point.
A retirement / shutdown / EOL date is when calls to the model stop working.
An intermediate date can matter too. On Amazon Bedrock the Legacy date is when new customers can no longer start using the model (and inactive existing customers may lose access), while the public extended access start date (for models with EOL dates after February 1, 2026) is when a Legacy model may move to higher provider-set pricing until EOL.

The same model can have different dates on different platforms

This is the single most important thing to know. A model offered through more than one platform is governed by each platform's own schedule, and those schedules diverge. Both Anthropic and Amazon Bedrock state this explicitly:

Anthropic: "The dates on this page apply to Anthropic-operated platforms ... Partner-operated platforms (Amazon Bedrock and Google Cloud) set their own retirement schedules, so a model's lifecycle status and dates can differ."

Amazon Bedrock: "Model lifecycle dates on this page are specific to Amazon Bedrock and may differ from dates published by model providers (such as Anthropic or Cohere). For Amazon Bedrock usage, only the dates on this page apply."

For example, Claude Sonnet 4 (claude-sonnet-4-20250514) is listed as Retired on the Anthropic API on June 15, 2026, while on Amazon Bedrock the same model (anthropic.claude-sonnet-4-20250514-v1:0) has an EOL date of October 14, 2026. Claude Opus 4.1 shows the same pattern: it was retired on the Anthropic API on August 5, 2026, while Amazon Bedrock lists an EOL date of January 8, 2027. Always check the row for the platform you actually call.

Sourcing

Every date and model ID below is transcribed from the vendor's official documentation at the time of writing, with the source page linked at the end of each vendor section. This page does not use second-hand reporting. Where an official page is dynamically rendered and its per-model table could not be transcribed verbatim, that is stated plainly and the reader is pointed to the official page. In the 2026-09-25 update, the OpenAI Deprecations page and the Vertex AI model-versions and deprecations pages, now published in the Gemini Enterprise Agent Platform documentation, were transcribed as well. In the 2026-10-10 update, the OpenAI, Anthropic, and Google tables were checked against the vendors' pages again.

Deprecation Calendar

Upcoming deadlines across all vendors

The consolidated table lists officially scheduled deadlines on or after this page's First Published date, in ascending order. Each platform's official lifecycle page is linked in the per-vendor sections below. Dates that have already passed are retained in the per-vendor tables for historical reference.

A date on this page only becomes actionable once you can answer the other half of the question, which is which of these model identifiers your own systems actually call. AI Bill of Materials on AWS covers that side — discovering the models, agents, and external endpoints in use on an AWS account rather than relying on someone remembering to write them down.

* You can sort the table by clicking on the column name.

Date	Platform	Model(s)	What happens	Replacement (as listed)
2026-07-05	Google Vertex AI	Claude 3.5 Haiku on Google Cloud	Shutdown	(not listed)
2026-07-07	Amazon Bedrock	Meta Llama 3.1 405B Instruct; Llama 3.2 1B / 3B / 11B / 90B Instruct	EOL	(not listed)
2026-07-09	Google Vertex AI	`gemini-2.5-flash-lite-preview-09-2025`; `gemini-2.5-flash-preview-05-2025`; `gemini-3.1-flash-lite-preview`	Retired	(not listed)
2026-07-17	Google Vertex AI	`gemini-3.1-flash-image-preview`; `gemini-3-pro-image-preview`	Retired	`gemini-3.1-flash-image` or `gemini-3-flash-image` (as written)
2026-07-23	OpenAI	14 legacy snapshots, including `computer-use-preview`; `gpt-5-codex`; `gpt-5.1-codex`; `gpt-5.1-codex-max`; `gpt-5.1-codex-mini`; `gpt-5.2-codex`; `o3-deep-research`; `o4-mini-deep-research`	Shutdown	`gpt-5.6-sol` / `gpt-5.6-terra` (per model)
2026-07-28	Google Gemini API	`gemini-2.5-computer-use-preview-10-2025`	Shutdown	`gemini-3.8-flash`
2026-07-30	Amazon Bedrock	Claude 3 Sonnet; Claude 3.5 Sonnet; Claude 3.5 Sonnet v2; Claude 3.7 Sonnet (GovCloud)	EOL	(not listed)
2026-08-05	Anthropic API	`claude-opus-4-1-20250805`	Retired	`claude-opus-4-8`
2026-08-10	OpenAI	`gpt-5.2-chat-latest`; `gpt-5.3-chat-latest`	Shutdown	`gpt-5.6-sol`
2026-08-10	Google Gemini API	`embedding-2-preview`	Shutdown	`gemini-embedding-2`
2026-08-17	Google Gemini API	`imagen-4.0-generate-001`; `imagen-4.0-ultra-generate-001`; `imagen-4.0-fast-generate-001`	Shutdown	`gemini-nano-banana-2.1`
2026-08-19	Amazon Bedrock	Cohere Command R; Command R+	EOL	(not listed)
2026-08-20	Google Vertex AI	Grok 4.1 family (`xai/grok-4.1-fast-reasoning`; `xai/grok-4.1-fast-non-reasoning`)	Shutdown	Newer xAI models, such as Grok 4.2 or Grok 4.3
2026-08-23	Google Vertex AI	Claude 3 Haiku on Google Cloud	Shutdown	(not listed)
2026-08-26	OpenAI	Assistants API (`/v1/assistants`, `/v1/threads`, and related endpoints)	Shutdown	Responses API and Conversations API
2026-08-31	Google Gemini API	`gemini-robotics-er-1.6-preview`	Shutdown	`gemini-robotics-er-2-preview`
2026-09-10	Amazon Bedrock	Claude 3 Haiku	EOL	(not listed)
2026-09-14	Amazon Bedrock	Amazon Nova Premier; Amazon Nova Sonic	EOL	(not listed)
2026-09-24	OpenAI	Videos API; `sora-2`; `sora-2-pro` (and dated snapshots)	Shutdown	(none listed)
2026-09-28	OpenAI	`gpt-3.5-turbo-instruct`; `babbage-002`; `davinci-002`; `gpt-3.5-turbo-1106`	Shutdown	`gpt-5.6-terra`
2026-09-30	Amazon Bedrock	Amazon Nova Canvas; Amazon Nova Reel (`v1:0`, `v1:1`)	EOL	(not listed)
2026-10-01	OpenAI	`gpt-5.4-cyber`	Shutdown	The most capable cyber model available to you
2026-10-05	Google Gemini API	`antigravity-preview-05-2026` (managed agent)	Shutdown	`antigravity-preview-09-2026`
2026-10-14	Amazon Bedrock	Claude Sonnet 4	EOL	(not listed)
2026-10-20	Google Vertex AI	`gemini-2.5-pro`; `gemini-2.5-flash`; `gemini-2.5-flash-lite`	Retirement	`gemini-3.8-flash` or `gemini-3.5-flash`; `gemini-3.8-flash`, `gemini-3.5-flash-lite`, or `gemini-3.1-flash-lite`; `gemini-3.8-flash`, `gemini-3.1-flash-lite`, or Gemma 4 (per model)
2026-10-21	Google Vertex AI	16 open-model MaaS endpoints (listed in the Vertex AI section)	Retirement	Self-deploy on Model Garden
2026-10-22	Google Gemini API	`gemini-omni-flash-preview`	Shutdown	`gemini-omni-1.1-flash`
2026-10-22	Google Gemini API	`veo-3.1-lite-generate-preview`; `veo-3.1-generate-preview`; `veo-3.1-fast-generate-preview`	Shutdown	`gemini-omni-1.1-flash`
2026-10-23	OpenAI	Legacy snapshots, including `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`; `o1-pro`; `o3-mini`; `o4-mini`, and fine-tuned models	Shutdown	`gpt-5.6-sol` / `gpt-5.6-terra` / `gpt-5.6-luna` / `gpt-image-2.5-sunburst` or `gpt-image-2.5-flare` (per model)
2026-10-23	Google Gemini API	`deep-research-pro-preview-12-2025` (managed agent)	Shutdown	`deep-research-preview-04-2026`
2026-11-17	Google Gemini API	`gemini-3.1-flash-tts-preview`; `gemini-2.5-flash-preview-tts`; `gemini-2.5-pro-preview-tts`	Shutdown	`gemini-3.8-flash-tts` or `gemini-3.8-flash-lite-tts`
2026-11-17	Google Gemini API	`gemini-3.1-flash-live-preview`; `gemini-2.5-flash-native-audio-preview-12-2025`	Shutdown	`gemini-3.8-live`
2026-11-19	Google Vertex AI	`gemini-3.6-flash`	Retirement	`gemini-3.8-flash`
2026-11-26	Amazon Bedrock	AI21 Jamba 1.5 Large; Jamba 1.5 Mini	EOL	(not listed)
2026-11-30	Anthropic API	`claude-sonnet-4-5-20250929`	Retirement	`claude-sonnet-5-5`
2026-11-30	Amazon Bedrock	TwelveLabs Marengo Embed v2.7	EOL	(not listed)
2026-11-30	OpenAI	`v1/prompts` API and reusable prompt objects; Evals dashboard and API; Agent Builder	Shutdown	(see the OpenAI section)
2026-12-01	OpenAI	`gpt-image-1-mini`; `gpt-image-1.5`; `chatgpt-image-latest`	Shutdown	`gpt-image-2.5-sunburst` or `gpt-image-2.5-flare`
2026-12-11	OpenAI	`gpt-5-2025-08-07`; `gpt-5-mini-2025-08-07`; `gpt-5-nano-2025-08-07`; `gpt-5-pro-2025-10-06`; `o3-2025-04-16`; `o3-pro-2025-06-10`	Shutdown	`gpt-5.6-sol` / `gpt-5.6-terra` / `gpt-5.6-luna` (per model)
2026-12-13	Google Vertex AI	`gemini-live-2.5-flash-native-audio`	Retirement	(not listed)
2027-01-06	OpenAI	Self-serve fine-tuning for active existing customers	New fine-tuning jobs can no longer be created	(not listed)
2027-01-06	OpenAI	`tts-1`; `tts-1-hd`; `gpt-4o-mini-tts-2025-03-20`; `gpt-4o-mini-tts-2025-12-15`	Shutdown	`gpt-realtime-2.1-mini`
2027-01-08	Amazon Bedrock	Claude Opus 4.1	EOL	(not listed)
2027-01-20	OpenAI	Legacy audio and realtime families (`gpt-realtime`; `gpt-audio`; `gpt-4o-audio`; `gpt-4o-realtime`, and their mini variants) and `gpt-4o-mini-transcribe-2025-03-20`	Shutdown	`gpt-realtime-2.1` / `gpt-realtime-2.1-mini` / `gpt-audio-1.5` (per model)
2027-01-28	Google Vertex AI	`gemini-3.7-flash`	Retirement	`gemini-3.8-flash`
2027-02-26	OpenAI	`whisper-1`; `gpt-4o-transcribe`; `gpt-4o-mini-transcribe`; `gpt-4o-transcribe-diarize`	Shutdown	`gpt-live-transcribe` or `gpt-transcribe`
2027-03-15	Google Gemini API	`gemini-2.5-flash-image`	Shutdown	`gemini-3.1-flash-lite-image`
2027-03-15	Google Vertex AI	`gemini-2.5-flash-image`	Retirement	`gemini-3.1-flash-lite-image`
2027-04-01	Google Vertex AI	`text-embedding-005`; `text-embedding-004`; `text-multilingual-embedding-002`; `multimodalembedding@001`	Retirement	(not listed)
2027-04-01	OpenAI	`gpt-5.3-codex`; `gpt-5.1`; `gpt-5.4-nano`	Shutdown	`gpt-6-sol` / `gpt-6-luna` (per model)
2027-04-08	Amazon Bedrock	Claude Sonnet 4.5	EOL	(not listed)
2027-05-07	Google Gemini API	`gemini-3.1-flash-lite`	Shutdown	`gemini-3.5-flash-lite`
2028-05-14	Google Gemini API	`gemini-embedding-001`	Shutdown	`gemini-embedding-2`

The per-vendor tables below give the full detail, including already-passed dates and the announcement or Legacy dates that precede each retirement.

Anthropic (Anthropic API / Claude Platform on AWS / Microsoft Foundry)

Anthropic assigns each model a lifecycle state (Active, Legacy, Deprecated, Retired) and, for deprecated models, a retirement date and a recommended replacement. For currently Active models, Anthropic lists a "Not sooner than" floor date rather than a scheduled retirement.

Currently Active models (earliest possible retirement):

Model ID	Retirement not sooner than
`claude-fable-5-1`	2027-09-01
`claude-mythos-5-1`	2027-09-01
`claude-fable-5`	2027-06-09
`claude-mythos-5`	2027-06-09
`claude-opus-5-5`	2027-09-22
`claude-opus-5`	2027-07-24
`claude-opus-4-8`	2027-05-28
`claude-opus-4-7`	2027-04-16
`claude-opus-4-6`	2027-02-05
`claude-opus-4-5-20251101`	2026-11-24
`claude-sonnet-5-5`	2027-09-28
`claude-sonnet-5`	2027-06-30
`claude-sonnet-4-6`	2027-02-17
`claude-haiku-5-5`	2027-10-07
`claude-haiku-4-5-20251001`	2026-10-15

In the first edition, claude-fable-5 and claude-sonnet-5 were not listed with a retirement floor. On 2026-09-25, the official model status table lists a floor for every Active model of the Claude 5 generation, and those rows are included above.

Deprecated and retired models (with retirement dates):

* You can sort the table by clicking on the column name.

Model ID	Deprecation announced	Retirement date	Recommended replacement	State
`claude-sonnet-4-5-20250929`	2026-09-30	2026-11-30	`claude-sonnet-5-5`	Deprecated
`claude-opus-4-1-20250805`	2026-06-05	2026-08-05	`claude-opus-4-8`	Retired
`claude-mythos-preview`	2026-06-09	To be announced	(not listed)	Deprecated
`claude-opus-4-20250514`	2026-04-14	2026-06-15	`claude-opus-4-8`	Retired
`claude-sonnet-4-20250514`	2026-04-14	2026-06-15	`claude-sonnet-5-5`	Retired
`claude-3-7-sonnet-20250219`	2025-10-28	2026-02-19	`claude-sonnet-5-5`	Retired
`claude-3-5-sonnet-20241022`	2025-08-13	2025-10-28	`claude-sonnet-5-5`	Retired
`claude-3-5-sonnet-20240620`	2025-08-13	2025-10-28	`claude-sonnet-5-5`	Retired
`claude-3-5-haiku-20241022`	2025-12-19	2026-02-19	`claude-haiku-4-5-20251001`	Retired
`claude-3-haiku-20240307`	2026-02-19	2026-04-20	`claude-haiku-4-5-20251001`	Retired
`claude-3-opus-20240229`	2025-06-30	2026-01-05	`claude-opus-4-8`	Retired
`claude-3-sonnet-20240229`	2025-01-21	2025-07-21	`claude-sonnet-5-5`	Retired
`claude-2.1`	2025-01-21	2025-07-21	`claude-opus-4-8`	Retired
`claude-2.0`	2025-01-21	2025-07-21	`claude-opus-4-8`	Retired
`claude-1.3`, `claude-1.2`, `claude-1.1`, `claude-1.0`	2024-09-04	2024-11-06	`claude-haiku-4-5-20251001`	Retired
`claude-instant-1.2`, `claude-instant-1.1`, `claude-instant-1.0`	2024-09-04	2024-11-06	`claude-haiku-4-5-20251001`	Retired

The claude-mythos-preview row follows the model status table on the official deprecations page, which on 2026-09-25 lists the model as Deprecated since June 9, 2026, with a retirement date to be announced. The first edition of this page recorded a retirement on June 30, 2026, taken from a notice on the same page; that notice no longer appears, and the model is not listed as retired.

Anthropic also lists an API-parameter deprecation: temperature, top_p, and top_k are deprecated on Claude 4.7 and later models, returning a 400 error when set to a non-default value on those models and on Claude Mythos Preview.

Official sources: Model deprecations - Claude Docs, Models overview - Claude Docs, Commitments on Model Deprecation and Preservation.

OpenAI

OpenAI's Deprecations page lists each deprecation with a shutdown date and a recommended replacement. The first edition of this page could not transcribe it; since the 2026-09-25 update, the table below transcribes every entry whose shutdown date falls on or after this page's First Published date (2026-07-05). Rows group the models that share an announcement date, a shutdown date, and a replacement. OpenAI's lifecycle terminology and notification practice are in the Vendor Lifecycle Policies section below.

* You can sort the table by clicking on the column name.

Announced	Shutdown date	Deprecated system	Recommended replacement
2026-04-22	2026-07-23	`computer-use-preview-2025-03-11` (`computer-use-preview`); `gpt-4o-mini-search-preview-2025-03-11`; `gpt-4o-search-preview-2025-03-11`; `gpt-5.1-codex-mini`	`gpt-5.6-terra`
2026-04-22	2026-07-23	`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`; `o3-deep-research-2025-06-26` (`o3-deep-research`); `o4-mini-deep-research-2025-06-26` (`o4-mini-deep-research`)	`gpt-5.6-sol`
2026-04-22	2026-07-23	`gpt-audio-mini-2025-10-06`	`gpt-audio-1.5`
2026-04-22	2026-07-23	`gpt-realtime-mini-2025-10-06`	`gpt-realtime-2.1-mini`
2026-05-08	2026-08-10	`gpt-5.2-chat-latest`; `gpt-5.3-chat-latest`	`gpt-5.6-sol`
2025-08-26	2026-08-26	Assistants API (`/v1/assistants`, `/v1/threads`, and related endpoints)	Responses API and Conversations API
2026-03-24	2026-09-24	Videos API; `sora-2`; `sora-2-pro`; `sora-2-2025-10-06`; `sora-2-2025-12-08`; `sora-2-pro-2025-10-06`	(none listed)
2025-09-26	2026-09-28	`gpt-3.5-turbo-instruct`; `babbage-002`; `davinci-002`; `gpt-3.5-turbo-1106`	`gpt-5.6-terra`
2026-09-11	2026-10-01	`gpt-5.4-cyber`	The most capable cyber model available to you
2026-04-22	2026-10-23	`gpt-3.5-turbo-0125` (`gpt-3.5-turbo`, `gpt-3.5-turbo-completions`); `o4-mini-2025-04-16` (`o4-mini`); `ft-o4-mini-2025-04-16`	`gpt-5.6-terra`
2026-04-22	2026-10-23	`gpt-4-0613` (`gpt-4`, `gpt-4-0613-completions`, `gpt-4-completions`); `gpt-4-1106-preview`; `gpt-4-turbo` (`gpt-4-turbo-2024-04-09`, `gpt-4-turbo-completions`); `gpt-4o-2024-05-13`; `o1-2024-12-17` (`o1`); `o3-mini-2025-01-31` (`o3-mini`)	`gpt-5.6-sol`
2026-04-22	2026-10-23	`o1-pro-2025-03-19` (`o1-pro`)	`gpt-5.6-sol` (`reasoning.mode: pro`)
2026-04-22	2026-10-23	`gpt-4.1-nano` (`gpt-4.1-nano-2025-04-14`)	`gpt-5.6-luna`
2026-04-22	2026-10-23	`gpt-image-1`	`gpt-image-2.5-sunburst` or `gpt-image-2.5-flare`
2026-04-22	2026-10-23	Fine-tuned `ft-gpt-3.5-turbo`; `ft-babbage-002`; `ft-davinci-002`	`gpt-5.6-terra` (base model)
2026-04-22	2026-10-23	Fine-tuned `ft-gpt-4`	`gpt-5.6-sol` (base model)
2026-04-22	2026-10-23	Fine-tuned `ft-gpt-4.1-nano-2025-04-14`	`gpt-5.6-luna` (base model)
2026-06-03	2026-11-30	Reusable prompt objects and the `v1/prompts` API	Move prompt content into application code
2026-06-03	2026-11-30	Evals platform (dashboard and API; existing evals become read-only on 2026-10-31)	Promptfoo (the migration path OpenAI documents)
2026-06-03	2026-11-30	Agent Builder	Agents SDK or ChatGPT Workspace Agents
2026-06-02	2026-12-01	`gpt-image-1-mini`; `gpt-image-1.5`; `chatgpt-image-latest`	`gpt-image-2.5-sunburst` or `gpt-image-2.5-flare`
2026-06-11	2026-12-11	`gpt-5-2025-08-07`; `o3-2025-04-16`	`gpt-5.6-sol`
2026-06-11	2026-12-11	`gpt-5-pro-2025-10-06`; `o3-pro-2025-06-10`	`gpt-5.6-sol` (`reasoning.mode: pro`)
2026-06-11	2026-12-11	`gpt-5-mini-2025-08-07`	`gpt-5.6-terra`
2026-06-11	2026-12-11	`gpt-5-nano-2025-08-07`	`gpt-5.6-luna`
2026-05-07	2027-01-06	Self-serve fine-tuning: active existing customers can no longer create new fine-tuning jobs (inference on fine-tuned models continues until the base model is deprecated)	(not listed)
2026-10-01	2027-01-06	`tts-1`; `tts-1-hd`; `gpt-4o-mini-tts-2025-03-20`; `gpt-4o-mini-tts-2025-12-15`	`gpt-realtime-2.1-mini`
2026-07-20	2027-01-20	`gpt-realtime`; `gpt-4o-realtime`	`gpt-realtime-2.1`
2026-07-20	2027-01-20	`gpt-realtime-mini`; `gpt-4o-mini-realtime`	`gpt-realtime-2.1-mini`
2026-07-20	2027-01-20	`gpt-audio`; `gpt-4o-audio`; `gpt-audio-mini`; `gpt-4o-mini-audio`	`gpt-audio-1.5`
2026-07-20	2027-01-20	`gpt-4o-mini-transcribe-2025-03-20`	`gpt-4o-mini-transcribe-2025-12-15`
2026-08-26	2027-02-26	`whisper-1`; `gpt-4o-transcribe`; `gpt-4o-mini-transcribe`; `gpt-4o-transcribe-diarize`	`gpt-live-transcribe` or `gpt-transcribe`
2026-10-01	2027-04-01	`gpt-5.3-codex`; `gpt-5.1`	`gpt-6-sol`
2026-10-01	2027-04-01	`gpt-5.4-nano`	`gpt-6-luna`

OpenAI's changelog records the Assistants API shutdown on 2026-08-26. For entries with earlier shutdown dates, and for the authoritative, current list, see OpenAI's official Deprecations page (linked below). Note the documentation host migration: the older platform.openai.com/docs/deprecations URL now redirects to developers.openai.com.

Official sources: OpenAI Deprecations, OpenAI Models.

Google — Gemini API (Google AI for Developers)

The Gemini API deprecations page publishes shutdown dates per model. Google states that "the shutdown dates listed in the table indicate the earliest possible dates on which a model might be retired," and that the exact date is communicated to users with advance notice. The table below lists models with a published shutdown date, in ascending order.

* You can sort the table by clicking on the column name.

Model ID	Shutdown date	Recommended replacement
`imagen-3.0-generate-002`	2025-11-10	`imagen-4.0-generate-001`
`gemini-2.0-flash-live-001`	2025-12-09	`gemini-3.8-live`
`gemini-live-2.5-flash-preview`	2025-12-09	`gemini-3.8-live`
`text-embedding-004`	2026-01-14	`gemini-embedding-2`
`gemini-3-pro-preview`	2026-03-09	`gemini-3.1-pro-preview`
`gemini-robotics-er-1.5-preview`	2026-04-30	`gemini-robotics-er-1.6-preview`
`gemini-3.1-flash-lite-preview`	2026-05-25	`gemini-3.1-flash-lite`
`gemini-2.0-flash`	2026-06-01	`gemini-3.6-flash`
`gemini-2.0-flash-001`	2026-06-01	`gemini-3.6-flash`
`gemini-2.0-flash-lite`	2026-06-01	`gemini-3.1-flash-lite`
`gemini-2.0-flash-lite-001`	2026-06-01	`gemini-3.1-flash-lite`
`gemini-3.1-flash-image-preview`	2026-06-25	`gemini-nano-banana-2.1`
`gemini-3-pro-image-preview`	2026-06-25	`gemini-3-pro-image`
`veo-3.0-generate-001`	2026-06-30	`veo-3.1-generate-preview` or the GA models on the Gemini Enterprise Agent Platform
`veo-3.0-fast-generate-001`	2026-06-30	`veo-3.1-fast-generate-preview` or the GA models on the Gemini Enterprise Agent Platform
`veo-2.0-generate-001`	2026-06-30	`veo-3.1-generate-preview` or the GA models on the Gemini Enterprise Agent Platform
`gemini-2.5-computer-use-preview-10-2025`	2026-07-28	`gemini-3.8-flash`
`embedding-2-preview`	2026-08-10	`gemini-embedding-2`
`imagen-4.0-generate-001`	2026-08-17	`gemini-nano-banana-2.1`
`imagen-4.0-ultra-generate-001`	2026-08-17	`gemini-nano-banana-2.1`
`imagen-4.0-fast-generate-001`	2026-08-17	`gemini-nano-banana-2.1`
`gemini-robotics-er-1.6-preview`	2026-08-31	`gemini-robotics-er-2-preview`
`antigravity-preview-05-2026` (managed agent)	2026-10-05	`antigravity-preview-09-2026`
`gemini-omni-flash-preview`	2026-10-22	`gemini-omni-1.1-flash`
`veo-3.1-lite-generate-preview`	2026-10-22	`gemini-omni-1.1-flash`
`veo-3.1-generate-preview`	2026-10-22	`gemini-omni-1.1-flash`
`veo-3.1-fast-generate-preview`	2026-10-22	`gemini-omni-1.1-flash`
`deep-research-pro-preview-12-2025` (managed agent)	2026-10-23	`deep-research-preview-04-2026`
`gemini-3.1-flash-tts-preview`	2026-11-17	`gemini-3.8-flash-tts` or `gemini-3.8-flash-lite-tts`
`gemini-2.5-flash-preview-tts`	2026-11-17	`gemini-3.8-flash-tts` or `gemini-3.8-flash-lite-tts`
`gemini-2.5-pro-preview-tts`	2026-11-17	`gemini-3.8-flash-tts` or `gemini-3.8-flash-lite-tts`
`gemini-3.1-flash-live-preview`	2026-11-17	`gemini-3.8-live`
`gemini-2.5-flash-native-audio-preview-12-2025`	2026-11-17	`gemini-3.8-live`
`gemini-2.5-flash-image`	2027-03-15	`gemini-3.1-flash-lite-image`
`gemini-3.1-flash-lite`	2027-05-07	`gemini-3.5-flash-lite`
`gemini-embedding-001`	2028-05-14	`gemini-embedding-2`

In the first edition, gemini-2.5-pro and gemini-2.5-flash were listed with a 2026-10-16 shutdown date, and gemini-embedding-001 with 2026-07-14. On 2026-09-25 the deprecations page shows "No shutdown date announced" for both 2.5 models and 2028-05-14 for gemini-embedding-001. The changelog entry of 2026-09-18 limits access to the 2.5 models to users who have actively used them before, and states that the models are not deprecated. Vertex AI sets its own dates for the same models (see the next section).

On 2026-10-10, the deprecations page (last updated 2026-10-09) shows a shutdown date of 2026-10-22 for gemini-omni-flash-preview, which this page listed on 2026-09-25 as 2026-09-30, and 2027-03-15 for gemini-2.5-flash-image, listed then as 2026-10-02 with gemini-3.1-flash-image-preview as the replacement; the replacement for the Imagen 4 models and gemini-3.1-flash-image-preview is now gemini-nano-banana-2.1. The changelog entry of 2026-10-06 deprecates gemini-3.1-flash-image with no shutdown date announced.

Google's changelog separately records the earlier Gemini 1.5 models (gemini-1.5-pro, gemini-1.5-flash, gemini-1.5-flash-8b) as shut down effective 2025-09-29; they no longer appear in the current deprecations table.

Official sources: Gemini API deprecations, Gemini API models.

Google — Vertex AI (Google Cloud)

Vertex AI (currently branded within the Gemini Enterprise Agent Platform documentation) maintains its own model-versioning and deprecations pages, separate from the Gemini API. In the first edition, its core "Model versions and lifecycle" page could not be transcribed verbatim. In the 2026-09-25 update, the page and the partner-model and open-model deprecations pages, now served from docs.cloud.google.com/gemini-enterprise-agent-platform, were readable, and the table below adds their entries dated on or after this page's First Published date:

* You can sort the table by clicking on the column name.

Item	Event	Date	Replacement / migration
`gemini-2.0-flash-001`, `gemini-2.0-flash-lite-001`	Discontinued	2026-06-01	Gemini 3.1 Flash-Lite, Gemma 4, or a more recent Gemini release
Vertex AI SDK "Generative AI module"	Deprecated 2025-06-24, shutdown 2026-06-24	2026-06-24	Google Gen AI SDK
Claude 3.5 Haiku on Google Cloud	Deprecated 2026-01-05, shutdown 2026-07-05	2026-07-05	(not listed)
`gemini-2.5-flash-lite-preview-09-2025`, `gemini-2.5-flash-preview-05-2025`, `gemini-3.1-flash-lite-preview`	Retired (release note of 2026-07-09)	2026-07-09	(not listed)
`gemini-3.1-flash-image-preview`, `gemini-3-pro-image-preview`	Retired (release note of 2026-07-17)	2026-07-17	`gemini-3.1-flash-image` or `gemini-3-flash-image` (as written)
Grok 4.1 family on the Gemini Enterprise Agent Platform (`xai/grok-4.1-fast-reasoning`, `xai/grok-4.1-fast-non-reasoning`)	Deprecated 2026-07-08, shutdown 2026-08-20	2026-08-20	Newer xAI models, such as Grok 4.2 or Grok 4.3
Claude 3 Haiku on Google Cloud	Deprecated 2026-02-23, shutdown 2026-08-23	2026-08-23	(not listed)
`gemini-2.5-pro`	Retirement	2026-10-20	`gemini-3.8-flash` or `gemini-3.5-flash`
`gemini-2.5-flash`	Retirement	2026-10-20	`gemini-3.8-flash`, `gemini-3.5-flash-lite`, or `gemini-3.1-flash-lite`
`gemini-2.5-flash-lite`	Retirement	2026-10-20	`gemini-3.8-flash`, `gemini-3.1-flash-lite`, or Gemma 4
16 open-model MaaS endpoints: `deepseek-ocr-maas`, `deepseek-r1-0528-maas`, `deepseek-v3.2-maas`, `deepseek-v3.1-maas`, `glm-5-maas`, `glm-4.7-maas`, `gpt-oss-20b-maas`, `kimi-k2-thinking-maas`, `llama-3.3-70b-instruct-maas`, `minimax-m2-maas`, `multilingual-e5-large-instruct-maas`, `multilingual-e5-small-maas`, `qwen3-235b-a22b-instruct-2507-maas`, `qwen3-coder-480b-a35b-instruct-maas`, `qwen3-next-80b-a3b-instruct-maas`, `qwen3-next-80b-a3b-thinking-maas`	Deprecated 2026-07-21, retirement 2026-10-21	2026-10-21	Self-deploy on Model Garden
`gemini-3.6-flash`	Retirement	2026-11-19	`gemini-3.8-flash`
`gemini-live-2.5-flash-native-audio`	Retirement	2026-12-13	(not listed)
`gemini-3.7-flash`	Retirement	2027-01-28	`gemini-3.8-flash`
`gemini-2.5-flash-image`	Retirement	2027-03-15	`gemini-3.1-flash-lite-image`
`text-embedding-005`, `text-embedding-004`, `text-multilingual-embedding-002`, `multimodalembedding@001`	Retirement	2027-04-01	(not listed)
`gemini-embedding-001`	Retirement no sooner than	2028-05-20	(not listed)

Vertex AI publishes stable model versions with retirement dates and auto-updated aliases that always point to the latest stable version. For the authoritative per-model retirement dates — including partner "Models as a Service" offerings such as Anthropic Claude and Mistral on Vertex AI — see the official pages linked below. The older cloud.google.com/vertex-ai deprecations page now carries a notice that it is no longer updated.

Official sources: Model versions and lifecycle - Gemini Enterprise Agent Platform, Partner model deprecations - Gemini Enterprise Agent Platform, Open model deprecations - Gemini Enterprise Agent Platform.

Amazon Bedrock

A model on Amazon Bedrock is in one of three states: Active, Legacy, or End-of-Life (EOL). The table below lists the models that the Legacy table of Amazon Bedrock has shown in the Legacy state or pending EOL since this page's First Published date; rows whose EOL date has passed are kept for reference. Since 2026-09-07, that table is on the Model lifecycle (Legacy) page, which applies to models launched before September 7, 2026; for models launched on or after that date, model cards show the dates where applicable (see Vendor Lifecycle Policies). On the current Model lifecycle (Legacy) page, models that have passed their EOL date are listed in a separate table ("These entries are retained for reference"), and the page does not list recommended replacements. The public extended access date (for EOL dates after February 1, 2026) marks when a Legacy model may move to higher, provider-set pricing until EOL.

* You can sort the table by clicking on the column name.

Provider	Model	Model ID	Legacy date	EOL date	Public extended access
Meta	Llama 3.1 405B Instruct	`meta.llama3-1-405b-instruct-v1:0`	2026-01-07	2026-07-07	2026-04-07
Meta	Llama 3.2 1B Instruct	`meta.llama3-2-1b-instruct-v1:0`	2026-01-07	2026-07-07	2026-04-07
Meta	Llama 3.2 3B Instruct	`meta.llama3-2-3b-instruct-v1:0`	2026-01-07	2026-07-07	2026-04-07
Meta	Llama 3.2 11B Instruct	`meta.llama3-2-11b-instruct-v1:0`	2026-01-07	2026-07-07	2026-04-07
Meta	Llama 3.2 90B Instruct	`meta.llama3-2-90b-instruct-v1:0`	2026-01-07	2026-07-07	2026-04-07
Anthropic	Claude 3 Sonnet	`anthropic.claude-3-sonnet-20240229-v1:0`	2026-01-30	2026-07-30	2026-04-30 (varies by Region)
Anthropic	Claude 3.5 Sonnet	`anthropic.claude-3-5-sonnet-20240620-v1:0`	2026-01-30	2026-07-30	2026-04-30
Anthropic	Claude 3.5 Sonnet v2	`anthropic.claude-3-5-sonnet-20241022-v2:0`	2026-01-30	2026-07-30	2026-04-30
Anthropic	Claude 3.7 Sonnet (GovCloud)	`anthropic.claude-3-7-sonnet-20250219-v1:0`	2026-01-30	2026-07-30	2026-04-30
Cohere	Command R	`cohere.command-r-v1:0`	2026-02-19	2026-08-19	2026-05-19
Cohere	Command R+	`cohere.command-r-plus-v1:0`	2026-02-19	2026-08-19	2026-05-19
Anthropic	Claude 3 Haiku	`anthropic.claude-3-haiku-20240307-v1:0`	2026-03-10	2026-09-10	2026-06-10
Amazon	Nova Premier	`amazon.nova-premier-v1:0`	2026-03-13	2026-09-14	(none)
Amazon	Nova Sonic	`amazon.nova-sonic-v1:0`	2026-03-13	2026-09-14	(none)
Amazon	Nova Canvas	`amazon.nova-canvas-v1:0`	2026-03-30	2026-09-30	(none)
Amazon	Nova Reel	`amazon.nova-reel-v1:0`	2026-03-30	2026-09-30	(none)
Amazon	Nova Reel	`amazon.nova-reel-v1:1`	2026-03-30	2026-09-30	(none)
Anthropic	Claude Sonnet 4	`anthropic.claude-sonnet-4-20250514-v1:0`	2026-04-14	2026-10-14	2026-07-14
AI21 Labs	Jamba 1.5 Large	`ai21.jamba-1-5-large-v1:0`	2026-05-26	2026-11-26	2026-08-26
AI21 Labs	Jamba 1.5 Mini	`ai21.jamba-1-5-mini-v1:0`	2026-05-26	2026-11-26	2026-08-26
TwelveLabs	Marengo Embed v2.7	`twelvelabs.marengo-embed-2-7-v1:0`	2026-05-29	2026-11-30	2026-08-29
Anthropic	Claude Opus 4.1	`anthropic.claude-opus-4-1-20250805-v1:0`	2026-07-08	2027-01-08	2026-10-08
Anthropic	Claude Sonnet 4.5	`anthropic.claude-sonnet-4-5-20250929-v1:0`	2026-10-08	2027-04-08	2027-01-08
Amazon	Titan Image Generator G1 v2	`amazon.titan-image-generator-v2:0`	2025-12-30	2026-06-30	(none)

Official sources: Model lifecycle (Legacy) - Amazon Bedrock User Guide, Model lifecycle - Amazon Bedrock User Guide.

Vendor Lifecycle Policies

This section summarizes each vendor's stated notification practice and model-ID conventions — the rules that govern how the dates above are set and communicated.

Anthropic

Notice: Anthropic "notifies customers with active deployments for models with upcoming retirements, providing at least 60 days notice before model retirement for publicly released models."
After retirement: "Requests to models past the retirement date will fail."
Model IDs carry a dated snapshot suffix (for example, claude-opus-4-5-20251101), and aliases (for example, claude-opus-4-5) point to the latest snapshot.
Scope: dates apply to Anthropic-operated platforms (the Claude API, Claude Platform on AWS, and Microsoft Foundry); Amazon Bedrock and Google Cloud set their own schedules.
Anthropic has published "Commitments on Model Deprecation and Preservation," including long-term preservation of model weights.

OpenAI

Notice: at least 6 months for generally available (GA) models; at least 3 months for specialized variants of GA models; preview models (whose names contain "preview") may be retired with much shorter notice, such as 2 weeks. Safety- or compliance-driven timelines can be shorter.
Notification method: impacted customers actively using a model are notified by email, with documentation updates and blog posts for larger changes.
Terminology: a model or endpoint becomes "deprecated" immediately when announced; at its "shutdown" (used interchangeably with "sunset") date it is "completely turned off" and "will no longer be accessible."

Google — Gemini API

Stability labels: stable ("usually don't change"; recommended for production), preview ("will be deprecated with at least 2 weeks notice"), experimental ("not ... suitable for production"), and latest (an alias that is "hot-swapped with every new release," with "a 2-week notice ... through email before the version behind latest is changed").
Shutdown dates in the table are "the earliest possible dates on which a model might be retired"; the exact date is communicated with advance notice.
At shutdown, a model is "completely turned off, and the endpoint is no longer available."

Google — Vertex AI

Vertex AI distinguishes latest stable models (recommended for new and active projects) from legacy stable models (older, still-supported versions), and marks a version retired once "past its retirement date and ... permanently deactivated."
Auto-updated aliases always point to the latest stable model; Google recommends pinning an explicit stable version for production.
Vertex AI's exact per-model retirement dates and notice windows are published on its official model-versions and deprecations pages (linked in the Vertex AI section above). The documentation host has migrated from cloud.google.com to docs.cloud.google.com, and the generative-AI docs are now branded under the "Gemini Enterprise Agent Platform."

Amazon Bedrock

Two policies by launch date (since 2026-09-07): the Model lifecycle page applies to models launched on Amazon Bedrock on or after September 7, 2026, and the Model lifecycle (Legacy) page applies to models launched before that date. For the newer models, model cards show, where applicable, an "EOL no sooner than" date and a Legacy period of either 6 months or 45 days (most models have 6 months; GPT-6.1 Sol follows OpenAI's first-party lifecycle terms and has no separate Bedrock minimum-availability date), and the EOL date is added to the model card when the Legacy period begins.
Minimum availability (models launched before 2026-09-07): "Once a model launches on Amazon Bedrock, it will remain on Amazon Bedrock for at least 12 months before the EOL date."
Legacy period (models launched before 2026-09-07): "A model will be in the Legacy state for at least 6 months before the EOL date." New customers cannot use Legacy models, and existing customers may lose access after 15 days of inactivity. You cannot create new Provisioned Throughput or new fine-tuning jobs on a Legacy model.
Customization: if you fine-tuned or customized a model before it entered Legacy, you can still create a new custom model deployment for on-demand inference and continue using existing on-demand deployments or Provisioned Throughput endpoints created before Legacy — but you cannot start new fine-tuning jobs or create new Provisioned Throughput endpoints once the model is Legacy.
Public extended access: for models with EOL dates after February 1, 2026, after at least 3 months in Legacy the model enters a public extended access period (a minimum of 3 more months to EOL) during which pricing may increase as set by the provider.
After EOL: "requests made to this version will fail, unless there is a private arrangement between you and the provider for continued access."
How it's surfaced: the state appears in the console and in the modelLifecycle field returned by the GetFoundationModel and ListFoundationModels API calls.

Frequently Asked Questions about AI Model Deprecations

What is the difference between deprecation and retirement?: Deprecation (or, on Amazon Bedrock, the Legacy state) means a model is still callable but is scheduled to go away: the vendor has flagged it, usually named a recommended replacement, and set a date. Retirement — called "shutdown" by OpenAI and the Gemini API, "retired" by Anthropic and Vertex AI, and "EOL" by Amazon Bedrock — is the date the model stops working. Before that date, existing code keeps running; after it, calls fail.
How do vendors announce model deprecations?: The notice periods differ. Anthropic provides at least 60 days notice before retirement for publicly released models. OpenAI gives at least 6 months for generally available models, at least 3 months for specialized variants, and as little as about 2 weeks for preview models. The Gemini API gives at least 2 weeks for preview and "latest" aliases and treats its table dates as the earliest possible shutdown. For models launched before 2026-09-07, Amazon Bedrock keeps a model at least 12 months on the platform and at least 6 months in the Legacy state before EOL; for models launched on or after that date, model cards state, where applicable, an "EOL no sooner than" date and a Legacy period of 6 months or 45 days. Vendors notify affected customers by email and update their documentation; larger changes are also announced on blogs.
What happens to API calls to a retired model?: They fail. Anthropic states that "requests to retired models will fail." Amazon Bedrock states that after EOL "requests made to this version will fail." The Gemini API states that a shut-down model is "completely turned off, and the endpoint is no longer available." OpenAI states that a shut-down model or endpoint "will no longer be accessible." Migration does not happen automatically — you must update your application code to a supported model before the date.
How should I plan a model migration?: Find your model's row for the specific platform you call (the same model can have different dates on different platforms), note the retirement, shutdown, or EOL date and the vendor's recommended replacement, test the replacement, and switch before the date. Step-by-step migration mechanics for Claude models are covered in the Anthropic Claude Model Migration Guide. To estimate context and token budgets for a candidate replacement, the LLM Token Counter and Context Budget Planner may help.

Summary

Model end-of-life is a recurring operational event across every AI platform, but each vendor publishes its schedule separately and uses different words — deprecated, legacy, retired, shutdown, sunset, EOL — for the same two underlying states. This page consolidates the officially published dates for the Anthropic API, OpenAI, Google (Gemini API and Vertex AI), and Amazon Bedrock, maps each vendor's vocabulary onto a common frame, and records that the same model can carry different dates on different platforms.

Since the 2026-09-25 update, this page transcribes the dated entries on the official lifecycle pages of the Anthropic API, OpenAI, the Google Gemini API, Google Vertex AI, and Amazon Bedrock, focusing on dates on or after its First Published date, together with each vendor's confirmed lifecycle policy. Further platforms may be added as warranted. The page is reviewed and updated quarterly and whenever major deprecations are announced; always confirm the current dates on the official pages linked in each section before acting.

Related references on hidekazu-konishi.com:

References:
Model deprecations - Claude Docs
OpenAI Deprecations
Gemini API deprecations
Model versions and lifecycle - Gemini Enterprise Agent Platform
Partner model deprecations - Gemini Enterprise Agent Platform
Open model deprecations - Gemini Enterprise Agent Platform
Model lifecycle - Amazon Bedrock User Guide
Model lifecycle (Legacy) - Amazon Bedrock User Guide

References:
Tech Blog with curated related content

Written by Hidekazu Konishi