Skip to content
The Model Graveyard

AI model deprecation tracker

The Model Graveyard

AI models are born, deprecated, and retired on schedules most teams never see coming. This is where they rest — with their retirement dates, live countdowns, and the successor you should migrate to before your app breaks.

Buried
21
On borrowed time
15
Still living
302
Immortal
19
Disabled
2
Next funeral
soon

Curated data verified · Methodology

Provider
Provider
Sort

Every AI model in one place — living, dying, and dead

The Model Graveyard is a live tracker of AI models from every major lab. Whether you are hunting for a specific LLM model, comparing options, or checking whether an AI model you depend on is being retired, this is the single list that stays current. We track hundreds of models, each with its release date, status, and — when the end comes — its retirement date and recommended replacement. It doubles as a stream of AI model news: new arrivals, scheduled deprecations, and silent removals, updated daily.

There are many types of AI models: chat and reasoning LLMs, generative AI models for images, video, audio, and code, plus embedding and moderation models. The graveyard covers the proprietary frontier and open source AI models alike. Browse the Claude models from Anthropic, the Gemini models from Google — including the current Google Gemini models and a full Gemini models list with their lifecycle — and the OpenAI models behind GPT and ChatGPT. Open-weight families such as Llama, Mistral, Qwen, and DeepSeek are marked “immortal,” because once weights are public a model can never truly be removed.

Compare, track, and avoid surprise retirements

Finding the best AI models is not a one-time decision — the top AI models change every few months. Use the search and filters to run a quick LLM model comparison by provider, status, or release date, and sort by what is newest or what is dying soonest. Looking for new AI model releases and announcements from the last few days? The News page lists every arrival in reverse-chronological order with an RSS feed, so you can follow launches as they happen.

The reason this site exists is the unglamorous part of the lifecycle: AI models deprecated and AI models removed on schedules that break production apps. Model IDs get hardcoded, then a provider sunsets the snapshot and every pinned request starts failing on the cutoff day. The Model Graveyard turns those retirement dates into live countdowns and tells you exactly which model to migrate to — the “survived by” successor each provider recommends. Status is computed from real dates, never hand-set, so a model labeled “on borrowed time” becomes “buried” automatically once its date passes. Treat this as the human-readable map of the whole field of AI models, and the official provider pages as the final word on any specific date.

AI model news, updated daily

The pace of releases makes AI model news hard to follow. Every LLM model seems to ship a new version monthly, and a list that is right today is wrong next week. That is why the catalog here is rebuilt every day from a live source: when a lab ships a model it appears as a new arrival, and when one quietly disappears it gets a headstone. If you are searching for new AI model releases announcements in the last 24 hours, the News page and RSS feed are the fastest way to see what just launched and what just died, without digging through a dozen changelogs.

Use it however you build. Pick the best AI models for a new feature by comparing the current Claude models, Gemini models, and open AI models side by side; audit an old codebase for any AI model that is already deprecated; or bookmark a single family — the Gemini models list, the open source AI models from Meta and Mistral, or OpenAI’s GPT line — and watch only what matters to you. However you use it, the goal is the same: never get surprised by a retired model again.

Frequently asked questions

Which AI is shutting down?

Several are on the way out. As of mid-2026, OpenAI is retiring GPT-4.5 and o3 from ChatGPT and shutting down the Assistants API, Google is retiring the Gemini 2.5 line on October 16, 2026, and Anthropic is retiring Claude Opus 4.1 on August 5, 2026. The graveyard above shows every model currently on borrowed time with a live countdown to its shutdown date.

What is a deprecated AI model?

A deprecated AI model is one the provider has flagged for retirement on a set date and discourages new use, even though it still works until that date. It is the warning window before shutdown: after the retirement date the model is removed and API requests to it fail. Deprecation is your cue to migrate before your app breaks.

Which new AI is launched recently?

New models ship almost weekly. Recent flagships include OpenAI's GPT-5.4, Anthropic's Claude Opus 4.8, and Google's Gemini 3.1 Pro, alongside fast-moving open-weight releases from Meta (Llama 4), Mistral, Alibaba (Qwen), and DeepSeek. See the News page for new AI model releases and announcements in reverse-chronological order, with an RSS feed.

What are the top 10 AI models?

The top AI models shift every few months, but as of mid-2026 the leading general-purpose models include GPT-5.4, GPT-5.4 Thinking, Claude Opus 4.8, Claude Sonnet 4.6, Gemini 3.1 Pro, Gemini 3 Flash, Llama 4, Mistral Large 3, DeepSeek-V3.2, and Qwen3-Max. Use the filters above to compare the current best AI models by provider and status.

What are some famous AI models?

Famous AI models include OpenAI's GPT series and ChatGPT, GPT-4o, and DALL·E; Anthropic's Claude; Google's Gemini; Meta's open-weight Llama; Mistral; and image models like Stable Diffusion. Many of the most famous — such as GPT-4o and DALL·E — are already buried in this graveyard, survived by newer successors.

What are common AI models?

Common production AI models are the widely deployed workhorses: GPT-5.4 and GPT-5.4 mini, Claude Sonnet 4.6 and Haiku 4.5, Gemini 3 Flash, and open models like Llama and Mistral for self-hosting, plus embedding models for search and retrieval. These are the models most apps actually call day to day.

What are the 4 models of AI?

In AI theory, the “four types of AI” usually means Reactive Machines, Limited Memory, Theory of Mind, and Self-Aware AI — a scale from simple reactive systems to hypothetical self-aware ones. Today's large language models, the ones tracked here, are Limited Memory systems. In everyday use, though, “AI models” means specific systems like GPT, Claude, or Gemini.

What's the difference between legacy, deprecated, and retired?

Legacy means an older model that still runs but is no longer the recommended default. Deprecated means a retirement date has been set and you should migrate before it. Retired (also called shut down or sunset) means the model is gone and requests to it return errors. The Model Graveyard maps these to Living, On borrowed time, and Buried.

When is GPT-4o being deprecated?

GPT-4o was retired from ChatGPT on April 3, 2026, after being the default for most of 2024 and 2025. OpenAI directs users to GPT-5.4 as the replacement. Always confirm exact dates on OpenAI’s official deprecations page, since API and ChatGPT timelines can differ.

Which Claude models are retiring in 2026?

Among others in 2026: Claude Opus 3 was retired on January 5, Claude 3.5 Sonnet, 3.5 Haiku and 3.7 Sonnet on February 19, Claude 3 Haiku on April 20, and Claude Sonnet 4 and Opus 4 on June 15. Claude Opus 4.1 is deprecated with a retirement date of August 5, 2026. Migrate toward Claude Opus 4.8, Sonnet 4.6, or Haiku 4.5.

When does Gemini 2.5 shut down?

Google has scheduled Gemini 2.5 Pro, 2.5 Flash, and 2.5 Flash-Lite to retire on October 16, 2026. The recommended path is the Gemini 3.x line — 3.1 Pro, 3 Flash, and 3.1 Flash-Lite. Vertex AI timelines can differ from the Gemini API, so check both.

Will my app break when a model is deprecated?

Your app keeps working through the deprecation window, but it will break on the retirement date if your code still calls the removed model ID — most providers return an error rather than silently routing to a newer model. Migrate before the date to avoid an outage.

Are deprecated models still usable?

Yes. A deprecated model still serves requests until its retirement date, so existing apps keep running during the window. After retirement it is no longer usable through the API. Occasionally a retired model lingers for paid or enterprise tiers, but you should not depend on that.

What should I migrate to when my model is deprecated?

Migrate to the successor the provider lists, which each memorial page records as "survived by". Generally that means moving to the current flagship or its same-tier equivalent — for example GPT-5.4, Claude Opus 4.8 or Sonnet 4.6, or Gemini 3.1 Pro — then re-testing prompts and outputs, since newer models can behave differently.

How do I find which model IDs my code uses?

Search your codebase and configuration for the provider’s ID strings — for example "gpt-", "claude-", or "gemini-" — including environment variables, prompt templates, and infrastructure-as-code. Centralize the model name in one config value so future migrations are a one-line change instead of a codebase-wide hunt.

How much notice do OpenAI, Anthropic, and Google give before retiring a model?

Notice varies by model and platform, often ranging from a couple of months to roughly a year, and is published on each provider’s deprecations page. Preview and snapshot models tend to get less notice than stable releases. Treat the official date as authoritative and subscribe to changelogs where possible.