When a provider retires a model, every request to that id starts failing, and the notice is easy to miss in a changelog. This tracker collects announced deprecations, retirement and shutdown dates and stated replacements for large language models from providers' own deprecation pages, changelogs and cloud lifecycle pages. It is for engineers, platform owners and AI governance teams who need to plan migrations before a shutdown, not after an outage. Providers are listed alphabetically; order is never a ranking. Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.
Key takeaways
- Several large shutdowns land before February 2027. OpenAI retires gpt-5, o3 and related models on 2026-12-11, Anthropic retires Claude Sonnet 4.5 on its own platforms on 2026-11-30, and Moonshot AI plans to retire the Kimi K2 series by the end of December 2026.
- Notice periods differ by provider and model type. Anthropic states at least 60 days for publicly released models; OpenAI states at least 6 months for GA models, 3 months for specialised and image models and as little as about 2 weeks for previews; Amazon Bedrock uses a 6 month or 45 day legacy period.
- Cloud platforms keep their own calendars. Anthropic says Amazon Bedrock and Google-managed Claude set their own schedules, and Bedrock retires Claude Sonnet 4.5 in April 2027 against 2026-11-30 on Anthropic's platforms.
- Redirects hide the change. Google redirected deprecated gemini-3.7-flash and gemini-3.5-flash requests to newer models, and DeepSeek routed retired V4 Flash ids to V4.1 Flash, so an id can keep answering with a different model.
- Retirement is not automatic migration. Amazon states migration does not happen automatically, and Anthropic states retired models return errors; inventory every caller before the date.
Status key
| Term | Meaning on this page |
|---|---|
| Deprecation notice | The date the provider announced the model will be retired. Deprecated models usually still answer; Anthropic describes them as less reliable and not recommended. |
| Retirement or shutdown | The date requests stop working (Anthropic: retired models return errors; Bedrock: EOL models fail in all Regions). |
| Redirect or routing | Requests to the old id are served by a newer model, as the provider states. |
| "Not sooner than" | The earliest possible retirement for an active model (Anthropic) or "or later" date (Google Cloud); not a scheduled shutdown. |
| Legacy (Bedrock) | Scheduled for retirement; no new customers, and existing customers may lose access after 15 days of inactivity. |
Announced deprecations and retirements
Retiring between 15 November 2026 and 31 January 2027
As of 2026-11-15. Providers alphabetical. Dates apply to the provider's own API unless marked.
| Provider | Model or id | Deprecation notice | Retirement or shutdown | Stated replacement | Source |
|---|---|---|---|---|---|
| Anthropic | Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) | 2026-09-30 | 2026-11-30 | claude-sonnet-5-5 | Anthropic deprecations |
| Anthropic | Claude Opus 4.8 fast mode (feature, not the model) | 2026-10-30 | 2026-11-30 | Not stated | Anthropic release notes |
| Gemini 3.6 Flash on Vertex (gemini-3.6-flash) | Not stated | 2026-11-19 (Vertex) | gemini-3.8-flash | Google Cloud model versions | |
| Gemini 3.7 Flash on Vertex (gemini-3.7-flash) | 2026-10-08 (Gemini API, redirected to gemini-3.8-flash) | 2027-01-28 (Vertex) | gemini-3.8-flash | Google Cloud model versions | |
| Moonshot AI | Kimi K2 series | Closed to new users from 2026-10-31 | Planned by end of December 2026 | Not stated | Kimi pricing page |
| OpenAI | /v1/prompts, Evals, Agent Builder (features) | 2026-06-03 | 2026-11-30 | Per feature (see source) | OpenAI deprecations |
| OpenAI | gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest | 2026-06-02 | 2026-12-01 | gpt-image-2.6-sunburst or gpt-image-2.6-flare | OpenAI deprecations |
| OpenAI | gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro | 2026-06-11 | 2026-12-11 | gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna (per model) | OpenAI deprecations |
| OpenAI | tts-1, tts-1-hd, gpt-4o-mini-tts | 2026-10-01 | 2027-01-06 | gpt-realtime-2.1-mini | OpenAI deprecations |
| OpenAI | GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini snapshots | 2026-10-14 | 2027-01-14 | gpt-6-luna | OpenAI deprecations |
| OpenAI | gpt-realtime, gpt-audio and gpt-4o realtime/audio variants | 2026-07-20 | 2027-01-20 | gpt-realtime-2.1 / gpt-audio-1.5 | OpenAI deprecations |
Scheduled for February 2027 onwards
As of 2026-11-15. Providers alphabetical. Dates apply to the provider's own API unless marked.
| Provider | Model or id | Deprecation notice | Retirement or shutdown | Stated replacement | Source |
|---|---|---|---|---|---|
| Mistral AI | Mistral Large 3 (mistral-large-2512) | 2026-10-31 | 2027-04-30 | Mistral Large 4 | Mistral models overview |
| Mistral AI | Mistral Medium 3.5 (mistral-medium-3-5) | 2026-10-31 | 2027-04-30 | Mistral Large 4 | Mistral models overview |
| Mistral AI | Voxtral TTS, Voxtral Mini Transcribe 2 and Realtime (2602 and 2603) | 2026-10-31 | 2027-04-30 | Voxtral 3 equivalents | Mistral models overview |
| OpenAI | gpt-5.5-cyber-preview | 2026-11-05 | 2027-02-01 | Most capable cyber model available to you | OpenAI deprecations |
| OpenAI | gpt-image-2 | 2026-11-10 | 2027-02-10 | gpt-image-2.6-sunburst or gpt-image-2.6-flare | OpenAI deprecations |
| OpenAI | whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe | 2026-08-26 | 2027-02-26 | gpt-live-transcribe or gpt-transcribe | OpenAI deprecations |
| OpenAI | gpt-5.1, gpt-5.3-codex | 2026-10-01 | 2027-04-01 | gpt-6-sol | OpenAI deprecations |
| OpenAI | gpt-5.4-nano | 2026-10-01 | 2027-04-01 | gpt-6-luna | OpenAI deprecations |
Retired or shut down in 2026
As of 2026-11-15. Providers alphabetical. Dates apply to the provider's own API unless marked.
| Provider | Model or id | Deprecation notice | Retirement or shutdown | Stated replacement | Source |
|---|---|---|---|---|---|
| Anthropic | Claude 3 Haiku (claude-3-haiku-20240307) | 2026-02-19 | 2026-04-20 | claude-haiku-4-5-20251001 | Anthropic deprecations |
| Anthropic | Claude Opus 4 and Claude Sonnet 4 (20250514) | 2026-04-14 | 2026-06-15 | claude-opus-4-8 / claude-sonnet-5-5 | Anthropic deprecations |
| Anthropic | Claude Opus 4.1 (claude-opus-4-1-20250805) | 2026-06-05 | 2026-08-05 | claude-opus-4-8 | Anthropic deprecations |
| Cohere | embed v2.0 models, c4ai-aya-expanse-8b, c4ai-aya-vision-8b | Not stated (effective date only) | 2026-04-04 | embed-v4.0 and Command models (see source) | Cohere deprecations |
| DeepSeek | DeepSeek-V4-Flash, V4-Flash-Vision-Exp | Not stated | 2026-09-10 (ids temporarily routed to V4.1 Flash) | deepseek-flash | DeepSeek API updates |
| DeepSeek | deepseek-chat, deepseek-reasoner (legacy names) | 2026-04-24 | Scheduled 2026-07-24 | deepseek-v4-pro / deepseek-v4-flash | DeepSeek API updates |
| gemini-3-pro-preview | 2026-02-26 | 2026-03-09 (redirects to gemini-3.1-pro-preview) | gemini-3.1-pro-preview | Gemini API changelog | |
| gemini-2.0-flash, gemini-2.0-flash-lite | 2026-02-18 | 2026-06-01 | Not stated | Gemini API changelog | |
| Gemini 2.5 Pro, Flash, Flash-Lite on Vertex | Not stated | 2026-10-20 (Vertex) | gemini-3.8-flash or gemini-3.5-flash / gemini-3.5-flash-lite | Google Cloud model versions | |
| Mistral AI | OCR 4.0, Leanstral 1.5 | 2026-09-29 | 2026-09-30 | OCR 4.1 / none listed | Mistral changelog |
| Mistral AI | Z.ai GLM 5.2 hosted (zai-glm-5-2) | 2026-09-29 | 2026-10-31 | Z.ai GLM 5.3 | Mistral models overview |
| OpenAI | gpt-3.5-turbo, gpt-4, gpt-4-turbo variants, o1, o3-mini, o4-mini and others | 2026-04-22 | 2026-10-23 | Mostly gpt-5.6-sol or gpt-5.6-terra | OpenAI deprecations |
| xAI | grok-voice-transcribe-1.0 | Not stated | 2026-10-02 (routes to grok-voice-transcribe-2.0) | grok-voice-transcribe-2.0 | xAI release notes |
| xAI | grok-imagine-image-quality, grok-imagine-image-pro | 2026-09-02 (image-quality); not stated (image-pro) | 2026-11-02 | grok-imagine-image-2.0-low | xAI release notes |
Amazon Bedrock schedules for hosted models
As of 2026-11-15, from the Amazon Bedrock model lifecycle (legacy) page. Bedrock sets its own dates, which can differ from the model provider's, and the page does not name replacements. Providers alphabetical.
| Provider | Model on Bedrock | Legacy date | EOL date |
|---|---|---|---|
| AI21 Labs | Jamba 1.5 Large and Mini | 2026-05-26 | 2026-11-26 |
| Amazon | Nova Canvas, Nova Reel | 2026-03-30 | 2026-09-30 (EOL reached) |
| Amazon | Nova Sonic (v1) | 2026-03-13 | 2026-09-14 (EOL reached) |
| Anthropic | Claude 3 Haiku, 3 Sonnet, 3.5 Sonnet, 3.5 Sonnet v2, 3.7 Sonnet | 2026-10-10 (Claude 3 Haiku and 3 Sonnet 2026-10-13 or 2026-10-15 in some Regions) | 2026-11-13 (Claude 3 Haiku and 3 Sonnet 2026-11-27 in some Regions) |
| Anthropic | Claude Opus 4.1 | 2026-07-08 | 2027-01-08 |
| Anthropic | Claude Sonnet 4.5 | 2026-10-08 or 2026-10-27 (by Region) | 2027-04-08 or 2027-04-27 (by Region) |
| Cohere | Command R, Command R+ | 2026-02-19 | 2026-08-19 (EOL reached) |
| DeepSeek | DeepSeek-R1, DeepSeek-V3.1 | 2026-09-30 | 2027-03-30 |
| Gemma 3 4B, 12B, 27B | 2026-09-30 | 2027-03-30 | |
| Meta | Llama 3, 3.1, 3.3 and Llama 4 Scout and Maverick | 2026-09-30 | 2027-03-30 |
| Mistral AI | Mistral 7B, Mixtral 8x7B, Devstral 2 123B, Voxtral 2507 | 2026-09-30 | 2027-03-30 |
| MiniMax | MiniMax M2, M2.1 | 2026-09-30 | 2027-03-30 |
| Moonshot AI | Kimi K2 Thinking | 2026-09-30 | 2027-03-30 |
| NVIDIA | Nemotron Nano 9B v2, Nemotron Nano 12B v2 VL | 2026-09-30 | 2027-03-30 |
| Qwen | Qwen3 Coder 30B A3B, Qwen3 Coder 480B A35B | 2026-09-30 | 2027-03-30 |
| Writer | Palmyra X4; Palmyra X5 | 2026-09-09; 2026-09-21 | 2027-03-09; 2027-03-21 |
| Z.AI | GLM 4.7, GLM 4.7 Flash | 2026-09-30 | 2027-03-30 |
Providers with no first-party model retirement notices on the pages checked
Meta's developer models page lists its Muse API ids without deprecation notices (Meta Llama models on Bedrock appear in the Bedrock table above). Not stated by Meta on the pages checked (2026-11-15). Alibaba Qwen, Baidu, ByteDance, MiniMax, Tencent and Zhipu Z.ai deprecation pages are not yet covered; providers not yet listed will be added at the next update.
Notice policies by provider
As of 2026-11-15. Policy text as each provider states it; providers alphabetical.
| Provider or platform | Stated minimum notice | Where notices appear |
|---|---|---|
| Amazon Bedrock | Legacy period of 6 months for most models or 45 days; EOL date added to the model card when Legacy starts | Model lifecycle pages, model cards, modelLifecycle field in ListFoundationModels |
| Anthropic | At least 60 days for publicly released models; email to customers with active deployments | Model deprecations page and documentation |
| Cohere | Not stated by Cohere on the page checked | Deprecations page |
| DeepSeek | Not stated by DeepSeek on the page checked (chat and reasoner names had three months) | API docs updates page |
| Not stated as a minimum on the pages checked; Google Cloud lists release and retirement dates per version | Gemini API changelog; Google Cloud model versions page | |
| Mistral AI | Not stated as a minimum on the pages checked; dated deprecation and retirement table per model | Models overview; changelog |
| Moonshot AI | Not stated by Moonshot AI on the page checked | Pricing page notice |
| OpenAI | At least 6 months for GA models, 3 months for specialised and image models, previews possibly about 2 weeks, unless safety or compliance needs require faster | Deprecations page |
| xAI | Not stated by xAI on the pages checked | Release notes and migration guides |
Parameter and feature removals that break requests
- Anthropic sampling parameters: temperature, top_p and top_k are removed in API version 2026-10-21 and requests that include them return a 400 error.
- Google sampling parameters: the Gemini API changelog marked temperature, top_p and top_k as deprecated on 2026-07-21.
- OpenAI features: /v1/prompts, the Evals dashboard and API, and Agent Builder shut down on 2026-11-30; /v1/audio/translations shuts down on 2027-02-26.
- Google access change: since 2026-09-18 Gemini 2.5 models on the Gemini API are limited to users who used them before; Google says they are not deprecated and are served until further notice.
What to watch next
- Claude Opus 4.5: Anthropic lists retirement not sooner than 2026-11-24, the earliest such date among its active models; Claude Haiku 4.5 (earliest date 2026-10-15) is still listed as Active with no deprecation date.
- Claude Mythos Preview: deprecated 2026-06-09 with retirement "to be announced".
- Gemini 2.5 on the Gemini API: served until further notice; Google recommends Gemini 3.5 Flash-Lite or 3.8 Flash for new projects.
- Bedrock end dates in late November 2026: Claude 3.x Sonnet models on 2026-11-13 to 27 by Region and AI21 Jamba 1.5 on 2026-11-26.
- gemini-3.1-flash-image: deprecated on 2026-10-06 with no shutdown date announced.
How to use this tracker
- Inventory callers. Search configuration, code and gateway logs for every model id in the tables, including ids reached through aliases. An AI Gateway and LLM Gateway Guide setup gives one place to see which ids are called.
- Use the right calendar. If you call a model through Bedrock or Google Cloud, use that platform's date, not the provider's.
- Test the replacement before the notice ends. Run your evaluation suite on the stated replacement, then switch in stages with a rollback id that is not itself on this list.
- Retire credentials with the model. Remove API keys, service accounts and gateway routes that only served the old model; see the AI Model Retirement Guide.
- Record the change. Log the migration through the AI Model Change Approval Template.
Practitioner checklist
- List every model id in production with provider, surface and the stated retirement or "not sooner than" date.
- Subscribe to each provider's deprecation page or changelog and to cloud lifecycle notices.
- Alert on any id within 90 days of its shutdown date.
- Check for redirects: confirm which model actually answers a deprecated id.
- Strip removed parameters (for example sampling parameters) before upgrading API versions.
- Revoke keys and routes used only by retired models.
- Check new adoptions against the LLM Release Tracker so you do not adopt a model that is already short lived.
Standards and references
- Anthropic: Model deprecations
- Amazon Web Services: Amazon Bedrock model lifecycle
- Cohere: Deprecations
- DeepSeek: API docs change log
- Google Cloud: Model versions and lifecycle
- Mistral AI: Models overview
- OpenAI: Deprecations
- xAI: Release notes
Related NHI Mgmt Group resources: LLM Release Tracker · LLM Pricing Tracker · Anthropic Models · OpenAI Models · Google Gemini Models