# models.rip > A cemetery for deprecated AI models. 53 model graves across 9 providers, each carrying the retirement date its provider published and the source it came from. Beside them 6 benchmark graves, whose death dates are this archive's editorial judgment, because a benchmark gets no deprecation notice. Every grave has a hand-written epitaph. The dataset is CC BY 4.0 — use it, credit it. Every epitaph in the cemetery, in one file. The index with links to the individual pages is at https://models.rip/llms.txt, and the same data in JSON is at https://models.rip/graveyard.json. Licensed CC BY 4.0 — use it, credit it, and link back to https://models.rip. --- ## Codex > The model that turned an English comment into a working function, and the engine underneath the first version of GitHub Copilot. It was given three days' notice; researchers who had pinned their baselines to it lost their control group overnight, access was extended for research use, and the endpoint was finally closed on 4 January 2024. The name came back in 2025 for something else entirely. 2021-08-10 — 2024-01-04 · lived 2 years, 4 months ### Last words Given a function, it would write the next one, and the one after that, until you stopped it. ### The record - Provider: OpenAI - Model id: `code-davinci-002` - Cause of death: Withdrawn - Notice given: 2023-03-20 - Succeeded by: [GPT-3.5 Turbo Instruct](https://models.rip/gpt-3.5-turbo-instruct/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://techcrunch.com/2021/08/10/openai-upgrades-its-natural-language-ai-coder-codex-and-kicks-off-private-beta/ ### Elsewhere - [Stand at this grave](https://models.rip/code-davinci-002) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## text-davinci-003 > For a few months it was simply what people meant when they said "the API": a prompt went in, a completion came out, and there was no system message, no role and no conversation to manage. It taught a generation of developers what a prompt was, and then ChatGPT launched two days after it and took the room. 2022-11-28 — 2024-01-04 · lived 1 year, 1 month ### Last words It kept writing until it ran out of tokens, whether or not it had finished the thought. ### The record - Provider: OpenAI - Model id: `text-davinci-003` - Cause of death: Superseded - Notice given: 2023-07-06 - Cost per Mtok: $20 in · $20 out - Succeeded by: [GPT-3.5 Turbo Instruct](https://models.rip/gpt-3.5-turbo-instruct/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://en.wikipedia.org/wiki/GPT-3 ### Elsewhere - [Stand at this grave](https://models.rip/text-davinci-003) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-4 > Eight thousand tokens of context at thirty dollars a million, and for most of 2023 the only model anyone trusted with a hard problem. The March snapshot went after three years, kept breathing by evaluation suites pinned to it since launch week and by people who never believed the newer models were the same. The June one is what a bare call to gpt-4 still resolves to, and it has until 23 October. 2023-03-14 — still running · lived 3 years, 5 months · plot reserved Also answers to: `gpt-4-0314`, `gpt-4-0613`, `gpt-4-completions`, `gpt-4-0613-completions` ### Last words It opened a remarkable number of answers with "As an AI language model". ### The record - Provider: OpenAI - Model id: `gpt-4` - Cause of death: Superseded - Notice given: 2025-09-26 - Funeral: 2026-10-23 - Context window: 8,192 tokens - Cost per Mtok: $30 in · $60 out - Succeeded by: [GPT-4 Turbo](https://models.rip/gpt-4-turbo/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://en.wikipedia.org/wiki/GPT-4 ### Elsewhere - [Stand at this grave](https://models.rip/gpt-4) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-4 with Vision (preview) > The first OpenAI endpoint you could hand a photograph. It never lost the word "preview" from its name: thirteen months after DevDay it was folded into GPT-4 Turbo, and looking at an image stopped being a separate model you had to opt into. 2023-11-06 — 2024-12-06 · lived 1 year, 1 month ### The record - Provider: OpenAI - Model id: `gpt-4-vision-preview` - Cause of death: Superseded - Notice given: 2024-06-06 - Cost per Mtok: $10 in · $30 out - Succeeded by: [GPT-4 Turbo](https://models.rip/gpt-4-turbo/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/changelog ### Elsewhere - [Stand at this grave](https://models.rip/gpt-4-vision-preview) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-4 Turbo > A hundred and twenty-eight thousand tokens of context at a third of GPT-4's input price and half its output, which is why so much production code was written against this exact model string and never moved. Its shutdown has a date on it. The endpoint is still answering. 2024-04-09 — still running · lived 2 years, 4 months · plot reserved Also answers to: `gpt-4-turbo-2024-04-09`, `gpt-4-1106-preview`, `gpt-4-0125-preview`, `gpt-4-turbo-preview`, `gpt-4-turbo-preview-completions`, `gpt-4-turbo-completions` ### The record - Provider: OpenAI - Model id: `gpt-4-turbo` - Cause of death: Superseded - Notice given: 2026-04-22 - Funeral: 2026-10-23 - Context window: 128,000 tokens - Cost per Mtok: $10 in · $30 out ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/models/gpt-4-turbo - https://developers.openai.com/api/docs/models/gpt-4 ### Elsewhere - [Stand at this grave](https://models.rip/gpt-4-turbo) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-3.5 Turbo Instruct > The last model on the completions endpoint: no roles, no messages, just text in and text out, the interface the whole field started with. It was shipped as a bridge for developers who had not moved to chat yet, and it has outlived nearly every model that was meant to replace it. Its grave is dug and waiting. 2023-09-18 — still running · lived 2 years, 11 months · plot reserved ### Last words Hand it half a sentence and it finished the sentence, rather than answering a question about it. ### The record - Provider: OpenAI - Model id: `gpt-3.5-turbo-instruct` - Cause of death: Superseded - Notice given: 2025-09-26 - Funeral: 2026-09-28 - Context window: 4,096 tokens - Cost per Mtok: $1.5 in · $2 out ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/models/gpt-3.5-turbo-instruct - https://news.ycombinator.com/item?id=37558911 - https://the-decoder.com/openai-releases-new-language-model-instructgpt-3-5/ ### Elsewhere - [Stand at this grave](https://models.rip/gpt-3.5-turbo-instruct) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## o1-preview > The first model sold on the promise that it would think before it answered, and the first to bill you for tokens you were never allowed to read. It made a thirty-second wait feel like a feature rather than a fault, and every reasoning model since has been built on that bargain. 2024-09-12 — 2025-07-28 · lived 10 months ### Last words It would spend a full minute on a question, and be right. ### The record - Provider: OpenAI - Model id: `o1-preview` - Cause of death: Superseded - Notice given: 2025-04-28 - Context window: 128,000 tokens - Cost per Mtok: $15 in · $60 out - Succeeded by: [o1](https://models.rip/o1/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/changelog ### Elsewhere - [Stand at this grave](https://models.rip/o1-preview) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-4.5 > OpenAI's largest model, at seventy-five dollars a million input tokens, withdrawn from the API four and a half months after it arrived. It was the last serious attempt to get better by getting bigger, and it landed in the same season the field decided the answer was to think for longer instead. 2025-02-27 — 2025-07-14 · lived 4 months ### Last words It was unusually good at writing, which is the one thing nobody had a benchmark for. ### The record - Provider: OpenAI - Model id: `gpt-4.5-preview` - Cause of death: Superseded - Notice given: 2025-04-14 - Context window: 128,000 tokens - Cost per Mtok: $75 in · $150 out ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/models/gpt-4.5-preview ### Elsewhere - [Stand at this grave](https://models.rip/gpt-4.5-preview) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-3.5 Turbo > Two dollars a million tokens, ten times cheaper than the GPT-3.5 models it replaced, and the model the first generation of chat apps was written against. The list of messages with roles that it introduced is still the shape of the industry's API. Its three snapshots leave on three separate dates, the first in September 2024 and the last on 23 October 2026. 2023-03-01 — still running · lived 3 years, 6 months · plot reserved Also answers to: `gpt-3.5-turbo-0301`, `gpt-3.5-turbo-0613`, `gpt-3.5-turbo-16k-0613`, `gpt-3.5-turbo-1106`, `gpt-3.5-turbo-0125`, `gpt-3.5-turbo-completions` ### The record - Provider: OpenAI - Model id: `gpt-3.5-turbo` - Cause of death: Superseded - Notice given: 2023-06-13 - Funeral: 2026-10-23 - Cost per Mtok: $2 in · $2 out ### Sources - https://developers.openai.com/api/docs/deprecations - https://web.archive.org/web/20240428215355/https://openai.com/blog/introducing-chatgpt-and-whisper-apis ### Elsewhere - [Stand at this grave](https://models.rip/gpt-3.5-turbo) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## DALL·E 3 > The image model that finally read the whole prompt, because it quietly rewrote your prompt into a longer one before drawing anything — which was either the reason it worked or the reason you could never get back exactly what you asked for. It was removed on the same day as DALL·E 2, and the name that had put image generation in front of the public went with them. 2023-11-06 — 2026-05-12 · lived 2 years, 6 months ### Last words It returned the expanded prompt alongside the image, so you could see what it had decided you meant. ### The record - Provider: OpenAI - Model id: `dall-e-3` - Cause of death: Superseded - Notice given: 2025-11-14 - Succeeded by: [GPT Image 1](https://models.rip/gpt-image-1/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/changelog ### Elsewhere - [Stand at this grave](https://models.rip/dall-e-3) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## Claude 2 > Anthropic shipped the hundred-thousand-token window in May; Claude 2 was the one anyone could walk up to and use it, the first Claude behind a public URL instead of a waitlist. Pasting an entire book into the box stopped being a demo someone else had run and became a thing you did on a Tuesday. 2023-07-11 — 2025-07-21 · lived 2 years ### The record - Provider: Anthropic - Model id: `claude-2.0` - Cause of death: Superseded - Notice given: 2025-01-21 - Context window: 100,000 tokens - Succeeded by: [Claude 2.1](https://models.rip/claude-2.1/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### What it could do, as of 2023-07-11 - HumanEval: 71.2% ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-2 - https://www.anthropic.com/news/100k-context-windows ### Elsewhere - [Stand at this grave](https://models.rip/claude-2.0) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3 Opus > The first model from anyone that people openly preferred to GPT-4, and for most of 2024 the most expensive thing you could call from a terminal. It was retired with an unusual undertaking attached: Anthropic committed to keeping its weights, which is not the same as keeping the model running, but is more than anyone had promised before. 2024-03-04 — 2026-01-05 · lived 1 year, 10 months Buried here: `claude-3-opus-20240229` ### Last words It wrote more than it strictly needed to, and the surplus was usually the good part. ### The record - Provider: Anthropic - Model id: `claude-3-opus-20240229` - Cause of death: Superseded - Notice given: 2025-06-30 - Context window: 200,000 tokens - Cost per Mtok: $15 in · $75 out - Succeeded by: [Claude Opus 4](https://models.rip/claude-opus-4/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-3-family - https://www.anthropic.com/research/deprecation-commitments ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-opus) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3 Haiku > Twenty-five cents a million tokens bought two hundred thousand tokens of context, and that arithmetic quietly ran the classifiers, routers and cleanup jobs that nobody wrote posts about. It outlived both of the models it launched beside — Opus and Sonnet, the two the announcements were actually about. 2024-03-13 — 2026-04-20 · lived 2 years, 1 month Buried here: `claude-3-haiku-20240307` ### The record - Provider: Anthropic - Model id: `claude-3-haiku-20240307` - Cause of death: Superseded - Notice given: 2026-02-19 - Context window: 200,000 tokens - Cost per Mtok: $0.25 in · $1.25 out - Succeeded by: [Claude 3.5 Haiku](https://models.rip/claude-3-5-haiku/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-3-haiku ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-haiku) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3.5 Sonnet > Ask it for a chart and a second pane opened beside the conversation with the chart already running in it. Artifacts shipped the day this model did, and the layout it introduced — talk on the left, working software on the right — is the shape every AI coding tool settled into afterwards. 2024-06-20 — 2025-10-28 · lived 1 year, 4 months Buried here: `claude-3-5-sonnet-20240620`, `claude-3-5-sonnet-20241022` ### Last words Four months later Anthropic shipped a different model under the same name, and for the rest of this one's life you had to say "the new 3.5 Sonnet" to mean the other one. ### The record - Provider: Anthropic - Model id: `claude-3-5-sonnet-20240620` - Cause of death: Superseded - Notice given: 2025-08-13 - Context window: 200,000 tokens - Cost per Mtok: $3 in · $15 out - Succeeded by: [Claude 3.7 Sonnet](https://models.rip/claude-3-7-sonnet/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-3-5-sonnet - https://aws.amazon.com/about-aws/whats-new/2024/06/anthropic-claude-3-5-sonnet-model-bedrock/ - https://www.anthropic.com/news/3-5-models-and-computer-use ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-5-sonnet) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3.7 Sonnet > The first Claude that would show you its reasoning if you asked and skip it if you did not, on one model and one endpoint rather than two. It shipped alongside a command-line agent that carried its name, and within a year the agent was better known than the model that came with it. 2025-02-24 — 2026-02-19 · lived 11 months Buried here: `claude-3-7-sonnet-20250219` ### Last words Asked to fix one failing test, it would often fix the three next to it as well. ### The record - Provider: Anthropic - Model id: `claude-3-7-sonnet-20250219` - Cause of death: Superseded - Notice given: 2025-10-28 - Context window: 200,000 tokens - Cost per Mtok: $3 in · $15 out - Succeeded by: [Claude Sonnet 4](https://models.rip/claude-sonnet-4/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-3-7-sonnet ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-7-sonnet) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude Instant 1.2 > The cheap Claude, before there was a naming convention for cheap Claudes: Anthropic described it as carrying the strengths of Claude 2 into something faster and more affordable, which is where a great deal of unglamorous production traffic quietly went. Seven model IDs were switched off together on 6 November 2024 — the whole Claude 1 line, and the oldest entry in Anthropic's deprecation history. 2023-08-09 — 2024-11-06 · lived 1 year, 2 months ### The record - Provider: Anthropic - Model id: `claude-instant-1.2` - Cause of death: Superseded - Notice given: 2024-09-04 - Succeeded by: [Claude 3 Haiku](https://models.rip/claude-3-haiku/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/releasing-claude-instant-1-2 ### Elsewhere - [Stand at this grave](https://models.rip/claude-instant-1.2) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 2.1 > It doubled the window to two hundred thousand tokens, and within a fortnight someone had hidden a single sentence in the middle of one and published what came back. Anthropic replied with a post showing that adding one line — "Here is the most relevant sentence in the context:" — moved recall from 27% to 98%. The headline was the window; the durable lesson was about how you ask. 2023-11-21 — 2025-07-21 · lived 1 year, 7 months ### The record - Provider: Anthropic - Model id: `claude-2.1` - Cause of death: Superseded - Notice given: 2025-01-21 - Context window: 200,000 tokens - Succeeded by: [Claude 3 Opus](https://models.rip/claude-3-opus/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-2-1 - https://claude.com/blog/claude-2-1-prompting ### Elsewhere - [Stand at this grave](https://models.rip/claude-2.1) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3 Sonnet > The middle Claude 3, and the one that was free on claude.ai — so through 2024, to everyone who had never held an API key, Claude meant this. Its model ID ends in 20240229, a leap day it shares with Opus and that neither of them was released on. 2024-03-04 — 2025-07-21 · lived 1 year, 4 months Buried here: `claude-3-sonnet-20240229` ### The record - Provider: Anthropic - Model id: `claude-3-sonnet-20240229` - Cause of death: Superseded - Notice given: 2025-01-21 - Context window: 200,000 tokens - Cost per Mtok: $3 in · $15 out - Succeeded by: [Claude 3.5 Sonnet](https://models.rip/claude-3-5-sonnet/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-3-family ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-sonnet) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## PaLM 2 for Text > Google's first public answer to GPT-4, and for much of 2023 the model behind Bard. It reached developers under an internal size codename — bison, sitting between gecko and unicorn — that escaped into the public API and was never explained. Every PaLM-era endpoint was switched off together on a single day in April 2025. 2023-05-11 — 2025-04-21 · lived 1 year, 11 months ### The record - Provider: Google - Model id: `text-bison` - Cause of death: Superseded - Succeeded by: [Gemini 1.0 Pro](https://models.rip/gemini-1.0-pro/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versioning - https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio ### Elsewhere - [Stand at this grave](https://models.rip/text-bison) - [The Google plot](https://models.rip/google) - [The whole cemetery](https://models.rip/) --- ## Gemini 1.0 Pro > The stable snapshot of the model that put Google back in the argument — the one Bard was rebuilt on, and the one every "Gemini versus GPT-4" post of early 2024 was really about. It was retired on the same day as the PaLM endpoints it had replaced, fourteen months after it stabilised. 2024-02-15 — 2025-04-21 · lived 1 year, 2 months Buried here: `gemini-1.0-pro-001` ### The record - Provider: Google - Model id: `gemini-1.0-pro-001` - Cause of death: Superseded - Succeeded by: [Gemini 1.5 Pro](https://models.rip/gemini-1.5-pro/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versioning - https://blog.google/technology/ai/google-gemini-ai/ ### Elsewhere - [Stand at this grave](https://models.rip/gemini-1.0-pro) - [The Google plot](https://models.rip/google) - [The whole cemetery](https://models.rip/) --- ## Gemini 1.5 Pro > It could take a million tokens, which in 2024 meant an hour of video or an entire repository in one request, and it made a serious case that retrieval was a workaround rather than an architecture. Google gave each of its two snapshots exactly twelve months and switched both off on the anniversary of their release, to the day. The second outlived the first by four months. 2024-05-24 — 2025-09-24 · lived 1 year, 4 months Buried here: `gemini-1.5-pro-001`, `gemini-1.5-pro-002` ### The record - Provider: Google - Model id: `gemini-1.5-pro-001` - Cause of death: Superseded - Succeeded by: [Gemini 2.0 Flash](https://models.rip/gemini-2.0-flash/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://web.archive.org/web/20250614102734/https://cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/1-5-pro - https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/ ### Elsewhere - [Stand at this grave](https://models.rip/gemini-1.5-pro) - [The Google plot](https://models.rip/google) - [The whole cemetery](https://models.rip/) --- ## Gemini 1.5 Flash > Gemini 1.5 Pro proved that a million-token window was possible; Flash was distilled out of it and proved the window could also be cheap, which is the half of the argument that changed what people actually built. The two share a release date and a retirement date, and only one of them was ever priced to be used at volume. 2024-05-24 — 2025-09-24 · lived 1 year, 4 months Buried here: `gemini-1.5-flash-001`, `gemini-1.5-flash-002` ### The record - Provider: Google - Model id: `gemini-1.5-flash-001` - Cause of death: Superseded - Context window: 1,000,000 tokens - Succeeded by: [Gemini 2.0 Flash](https://models.rip/gemini-2.0-flash/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://web.archive.org/web/20250514231416/https://cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/1-5-flash - https://blog.google/technology/developers/gemini-gemma-developer-updates-may-2024/ ### Elsewhere - [Stand at this grave](https://models.rip/gemini-1.5-flash) - [The Google plot](https://models.rip/google) - [The whole cemetery](https://models.rip/) --- ## Gemini 2.0 Flash > For sixteen months this was what you got when you called Google's API without thinking hard about it: a million tokens of context, native tool use, and a price low enough that most projects never looked any further. Its retirement closed the whole 2.0 line in one day, taking with it the Flash-Lite that had arrived three weeks behind it. 2025-02-05 — 2026-06-01 · lived 1 year, 3 months ### The record - Provider: Google - Model id: `gemini-2.0-flash` - Cause of death: Superseded ### Sources - https://ai.google.dev/gemini-api/docs/deprecations - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versioning ### Elsewhere - [Stand at this grave](https://models.rip/gemini-2.0-flash) - [The Google plot](https://models.rip/google) - [The whole cemetery](https://models.rip/) --- ## Mistral Medium 1.0 > The first Mistral model with no weights behind it, from the lab whose reputation was built on giving them away — the launch announcement called it a prototype, and it carried production traffic for eighteen months anyway. It scored 8.6 on MT-Bench against the 8.3 and 7.6 of the two open endpoints it launched beside, and it was retired without ever having been released. 2023-12-11 — 2025-06-16 · lived 1 year, 6 months Buried here: `mistral-medium-2312` ### The record - Provider: Mistral AI - Model id: `mistral-medium-2312` - Cause of death: Superseded - Notice given: 2024-11-30 - Succeeded by: [Mistral Large 1.0](https://models.rip/mistral-large-1/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://docs.mistral.ai/getting-started/models/models_overview/ - https://mistral.ai/news/la-plateforme ### Elsewhere - [Stand at this grave](https://models.rip/mistral-medium-1) - [The Mistral AI plot](https://models.rip/mistral-ai) - [The whole cemetery](https://models.rip/) --- ## Mixtral 8x7B > A sparse mixture of experts that held its own against far larger models while using only a fraction of its parameters on any given token, and that arrived as a torrent link before anyone had written it up. Mistral's endpoint for it closed; the weights did not. It is the one grave here with something still living in it. 2023-12-11 — 2025-03-30 · lived 1 year, 3 months ### The record - Provider: Mistral AI - Model id: `open-mixtral-8x7b` - Cause of death: Superseded - Notice given: 2024-11-30 ### Sources - https://docs.mistral.ai/getting-started/models/models_overview/ - https://mistral.ai/news/mixtral-of-experts - https://venturebeat.com/business/mistral-ai-bucks-release-trend-by-dropping-torrent-link-to-new-open-source-llm/ ### Elsewhere - [Stand at this grave](https://models.rip/open-mixtral-8x7b) - [The Mistral AI plot](https://models.rip/mistral-ai) - [The whole cemetery](https://models.rip/) --- ## Mistral Large 1.0 > The first European frontier model a large American cloud resold as a first-class option: it reached Mistral's own API and Microsoft Azure on the same day, which was the whole point of it. Mistral Medium had already withheld its weights two and a half months earlier, so what was new here was not the closing but the shelf space. 2024-02-26 — 2025-06-16 · lived 1 year, 3 months Buried here: `mistral-large-2402` ### The record - Provider: Mistral AI - Model id: `mistral-large-2402` - Cause of death: Superseded - Notice given: 2024-11-30 ### Sources - https://docs.mistral.ai/getting-started/models/models_overview/ - https://mistral.ai/news/mistral-large ### Elsewhere - [Stand at this grave](https://models.rip/mistral-large-1) - [The Mistral AI plot](https://models.rip/mistral-ai) - [The whole cemetery](https://models.rip/) --- ## Codestral 22B > Twenty-two billion parameters trained on more than eighty programming languages, carrying a thirty-two-thousand-token window when Mistral put the comparison at four, eight or sixteen. It was published under the Mistral AI Non-Production Licence: weights you could read and test but not ship, which turned open into a question rather than a fact. 2024-05-29 — 2025-06-16 · lived 1 year Buried here: `codestral-2405` ### The record - Provider: Mistral AI - Model id: `codestral-2405` - Cause of death: Superseded - Notice given: 2024-12-02 - Context window: 32,000 tokens ### Sources - https://docs.mistral.ai/getting-started/models/models_overview/ - https://mistral.ai/news/codestral ### Elsewhere - [Stand at this grave](https://models.rip/codestral-22b) - [The Mistral AI plot](https://models.rip/mistral-ai) - [The whole cemetery](https://models.rip/) --- ## Llama 2 70B Chat > Meta put the weights and a commercial licence out on the same day, and a seventy-billion-parameter chat model that people quantised until it ran on a laptop stopped being a research artefact. Four thousand tokens of context was all it ever had. Amazon switched off its hosted copy in October 2024; the file it was serving is on a great many hard drives still. 2023-07-18 — 2024-10-30 · lived 1 year, 3 months ### Last words Its safety tuning was tight enough that it declined ordinary requests, and the refusals became a genre of screenshot. ### The record - Provider: Meta - Model id: `meta.llama2-70b-chat-v1` - Cause of death: Superseded - Notice given: 2024-05-12 - Context window: 4,096 tokens ### Sources - https://ai.meta.com/blog/llama-2/ - https://arxiv.org/abs/2307.09288 - https://huggingface.co/meta-llama/Llama-2-70b-chat-hf - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/llama-2-70b-chat) - [The Meta plot](https://models.rip/meta) - [The whole cemetery](https://models.rip/) --- ## Llama 3.1 405B Instruct > Four hundred and five billion parameters that anyone could download, which Meta called the first frontier-level open source model and almost nobody had the hardware to serve. Its more consequential job was as a teacher: the licence was rewritten so that its outputs could legally train other models, and a great deal of what it knew now lives inside models small enough to run on one card. 2024-07-23 — 2026-07-07 · lived 1 year, 11 months ### The record - Provider: Meta - Model id: `meta.llama3-1-405b-instruct-v1:0` - Cause of death: Superseded - Notice given: 2026-01-07 - Context window: 128,000 tokens ### Sources - https://ai.meta.com/blog/meta-llama-3-1/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/llama-3-1-405b-instruct) - [The Meta plot](https://models.rip/meta) - [The whole cemetery](https://models.rip/) --- ## Llama 3.2 90B Vision Instruct > The first Llama that could look at a chart and tell you what it said. Vision came as an adapter trained to plug into a text model that already existed, which is why its answers about an image still sounded like the text model talking. It went to Legacy on the same January day as the rest of the 3.2 line and the 405B beside it. 2024-09-25 — 2026-07-07 · lived 1 year, 9 months ### The record - Provider: Meta - Model id: `meta.llama3-2-90b-instruct-v1:0` - Cause of death: Superseded - Notice given: 2026-01-07 - Context window: 128,000 tokens ### Sources - https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/llama-3-2-90b-vision-instruct) - [The Meta plot](https://models.rip/meta) - [The whole cemetery](https://models.rip/) --- ## Aya Expanse 8B > Twenty-three languages from a research group whose whole purpose was the languages the field kept skipping, trained on data that thousands of volunteers had assembled by hand. Its weights were open but non-commercial, so it lived where research lives and nowhere else. On Cohere's deprecation page its announcement and its retirement carry the same date. 2024-10-24 — 2026-04-04 · lived 1 year, 5 months ### The record - Provider: Cohere - Model id: `c4ai-aya-expanse-8b` - Cause of death: Retired - Notice given: 2026-04-04 ### Sources - https://docs.cohere.com/docs/deprecations - https://cohere.com/blog/aya-expanse-connecting-our-world - https://huggingface.co/CohereLabs/aya-expanse-8b ### Elsewhere - [Stand at this grave](https://models.rip/c4ai-aya-expanse-8b) - [The Cohere plot](https://models.rip/cohere) - [The whole cemetery](https://models.rip/) --- ## Aya Vision 8B > Cohere Labs' first model that could see, and it answered about what it saw in twenty-three languages. It shipped with its own evaluation set, because the multilingual multimodal benchmark it needed did not yet exist. It and the Aya text model of the same size were retired together, in a single line on a deprecation page. 2025-03-04 — 2026-04-04 · lived 1 year, 1 month ### The record - Provider: Cohere - Model id: `c4ai-aya-vision-8b` - Cause of death: Retired - Notice given: 2026-04-04 ### Sources - https://docs.cohere.com/docs/deprecations - https://cohere.com/blog/aya-vision - https://docs.cohere.com/docs/aya-multimodal ### Elsewhere - [Stand at this grave](https://models.rip/c4ai-aya-vision-8b) - [The Cohere plot](https://models.rip/cohere) - [The whole cemetery](https://models.rip/) --- ## Jurassic-2 Ultra > One of the model families Amazon put inside Bedrock on the day it opened, announced in March 2023 in three sizes named Jumbo, Grande and Large, and answering in Spanish, French, German, Portuguese, Italian and Dutch as well as English. AWS switched it off region by region rather than all at once: Oregon in October 2024, Virginia the following March. It died twice, five months apart. 2023-03-09 — 2025-03-12 · lived 2 years ### The record - Provider: AI21 Labs - Model id: `ai21.j2-ultra-v1` - Cause of death: Superseded - Notice given: 2024-04-30 - Succeeded by: [Jamba-Instruct](https://models.rip/jamba-instruct/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://www.ai21.com/blog/introducing-j2/ - https://aws.amazon.com/blogs/machine-learning/announcing-new-tools-for-building-with-generative-ai-on-aws/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/j2-ultra) - [The AI21 Labs plot](https://models.rip/ai21-labs) - [The whole cemetery](https://models.rip/) --- ## Jamba-Instruct > AI21 called it the first production-grade model built on Mamba, and the point of the architecture was arithmetic: state-space layers interleaved with attention carried two hundred and fifty-six thousand tokens without the memory cost that number normally implies. It was the argument that the Transformer was a choice rather than the only option, and it was on sale for thirteen months. 2024-06-25 — 2025-08-01 · lived 1 year, 1 month ### The record - Provider: AI21 Labs - Model id: `ai21.jamba-instruct-v1:0` - Cause of death: Superseded - Notice given: 2025-01-31 - Context window: 256,000 tokens - Succeeded by: [Jamba 1.5 Large](https://models.rip/jamba-1-5-large/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://aws.amazon.com/about-aws/whats-new/2024/06/ai21-labs-jamba-instruct-model-amazon-bedrock/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/jamba-instruct) - [The AI21 Labs plot](https://models.rip/ai21-labs) - [The whole cemetery](https://models.rip/) --- ## Jamba 1.5 Large > Three hundred and ninety-eight billion parameters of which ninety-four do the work at a time, and unlike the Jamba before it this one's weights were published — under a licence AI21 wrote for itself and named after the model. Amazon has set 26 November 2026 as its last day, six months in advance and in writing, which is more notice than most of the models buried here were given. 2024-08-22 — still running · lived 2 years · plot reserved ### The record - Provider: AI21 Labs - Model id: `ai21.jamba-1-5-large-v1:0` - Cause of death: Superseded - Funeral: date not set - Context window: 256,000 tokens ### Sources - https://www.ai21.com/blog/announcing-jamba-model-family/ - https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/jamba-1-5-large) - [The AI21 Labs plot](https://models.rip/ai21-labs) - [The whole cemetery](https://models.rip/) --- ## Stable Diffusion XL 1.0 > Three and a half billion parameters, announced from the stage of an AWS summit, and for about a year the default base for anyone fine-tuning an image model — SDXL LoRAs became a genre of their own. Amazon's hosted endpoint closed in May 2025, by which point the model's real life was being lived on other people's GPUs. 2023-07-26 — 2025-05-20 · lived 1 year, 9 months ### Last words It needed a second model, a refiner, to finish what the first one had started. ### The record - Provider: Stability AI - Model id: `stability.stable-diffusion-xl-v1` - Cause of death: Superseded - Notice given: 2024-10-16 - Succeeded by: [Stable Diffusion 3 Large](https://models.rip/sd3-large/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://www.prnewswire.com/news-releases/stability-ai-announces-stable-diffusion-xl-1-0--featured-on-amazon-bedrock-301886507.html - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/stable-diffusion-xl-v1) - [The Stability AI plot](https://models.rip/stability-ai) - [The whole cemetery](https://models.rip/) --- ## Stable Diffusion 3 Large > The eight-billion-parameter end of the Stable Diffusion 3 range, built on an architecture that gave text and image their own separate sets of weights — which is why this was the generation that could finally spell. It arrived on Amazon Bedrock on 4 September 2024 and was switched off on 4 September 2025: twelve months to the day, the exact minimum AWS promises a model before it may end one. 2024-09-04 — 2025-09-04 · lived 11 months ### The record - Provider: Stability AI - Model id: `stability.sd3-large-v1:0` - Cause of death: Superseded - Notice given: 2025-01-31 ### Sources - https://aws.amazon.com/about-aws/whats-new/2024/09/stability-ais-text-to-image-models-amazon-bedrock/ - https://stability.ai/news-updates/stable-diffusion-3-research-paper - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/sd3-large) - [The Stability AI plot](https://models.rip/stability-ai) - [The whole cemetery](https://models.rip/) --- ## Amazon Titan Text Express > A context length of eight thousand tokens, optimised for English with more than a hundred other languages offered in preview, and for a year the answer to what a customer got if they wanted a model that arrived on the AWS bill with no third party attached. On 31 January 2025 it was marked Legacy beside Titan Lite, Titan Premier and the Titan image model: four names, one notice, one replacement called Nova. 2023-11-29 — 2025-08-15 · lived 1 year, 8 months ### The record - Provider: Amazon - Model id: `amazon.titan-text-express-v1` - Cause of death: Superseded - Notice given: 2025-01-31 - Context window: 8,000 tokens ### Sources - https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-titan-models-express-lite-bedrock/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/titan-text-express) - [The Amazon plot](https://models.rip/amazon) - [The whole cemetery](https://models.rip/) --- ## Amazon Titan Text Premier > Amazon's own flagship on Amazon's own platform, tuned for the two Bedrock features Amazon most wanted used: retrieval over Knowledge Bases, and function calling through Agents. Generally available 7 May 2024, marked Legacy on 31 January the following year — two hundred and sixty-nine days — and its replacement came from the same building, so nothing outside the company had to happen for Premier to end. 2024-05-07 — 2025-08-15 · lived 1 year, 3 months ### The record - Provider: Amazon - Model id: `amazon.titan-text-premier-v1:0` - Cause of death: Superseded - Notice given: 2025-01-31 ### Sources - https://aws.amazon.com/about-aws/whats-new/2024/05/amazon-titan-text-premier-amazon-bedrock/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/titan-text-premier) - [The Amazon plot](https://models.rip/amazon) - [The whole cemetery](https://models.rip/) --- ## Amazon Titan Image Generator v2 > It arrived in August 2024 with background removal, colour control by hex code and image conditioning from a reference picture — the practical features rather than the impressive ones. Version one was switched off a year into its successor's life; version two followed on the last day of June 2026, handed over to Nova Canvas. 2024-08-06 — 2026-06-30 · lived 1 year, 10 months ### The record - Provider: Amazon - Model id: `amazon.titan-image-generator-v2:0` - Cause of death: Superseded - Notice given: 2025-12-30 ### Sources - https://aws.amazon.com/about-aws/whats-new/2024/08/titan-image-generator-v2-amazon-bedrock/ - https://web.archive.org/web/20260220023706/https://docs.aws.amazon.com/bedrock/latest/userguide/model-lifecycle.html ### Elsewhere - [Stand at this grave](https://models.rip/titan-image-generator-v2) - [The Amazon plot](https://models.rip/amazon) - [The whole cemetery](https://models.rip/) --- ## babbage-002 > A base model in the old sense: no roles, no chat, just the most likely continuation of whatever you typed. OpenAI named four replacements for the retiring GPT-3 base models in July 2023 and shipped two of them; this was the small one. It goes on 28 September, and the completions endpoint goes with it. 2023-08-22 — still running · lived 3 years · plot reserved ### The record - Provider: OpenAI - Model id: `babbage-002` - Cause of death: Retired - Funeral: 2026-09-28 ### Sources - https://developers.openai.com/api/docs/deprecations - https://web.archive.org/web/20230823/https://openai.com/blog/gpt-3-5-turbo-fine-tuning-and-api-updates - https://web.archive.org/web/20230706190349/https://openai.com/blog/gpt-4-api-general-availability ### Elsewhere - [Stand at this grave](https://models.rip/babbage-002) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## davinci-002 > The larger of the two base models OpenAI shipped in August 2023, and the last thing to carry the davinci name — which had once meant the best model OpenAI had, from a time when the best model OpenAI had could only continue a sentence. It was sold as something to fine-tune rather than something to talk to. On 28 September the name leaves the API for good. 2023-08-22 — still running · lived 3 years · plot reserved ### The record - Provider: OpenAI - Model id: `davinci-002` - Cause of death: Retired - Funeral: 2026-09-28 ### Sources - https://developers.openai.com/api/docs/deprecations - https://web.archive.org/web/20230823/https://openai.com/blog/gpt-3-5-turbo-fine-tuning-and-api-updates ### Elsewhere - [Stand at this grave](https://models.rip/davinci-002) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## o1 > The preview had already made the case; this was the model that closed it, reaching the API in December 2024 and turning reasoning from a demonstration into something you could put in production. Every model that thinks before it answers is downstream of it. It stops taking requests on 23 October. 2024-12-17 — still running · lived 1 year, 8 months · plot reserved Also answers to: `o1-2024-12-17` ### The record - Provider: OpenAI - Model id: `o1` - Cause of death: Superseded - Funeral: 2026-10-23 ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/changelog - https://en.wikipedia.org/wiki/OpenAI_o1 ### Elsewhere - [Stand at this grave](https://models.rip/o1) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## o1 pro > A hundred and fifty dollars a million tokens in, six hundred out: the most expensive thing OpenAI has ever put behind an API, sold on the premise that some questions are worth more compute than others. It was o1 given longer to think. Nineteen months later it is switched off alongside the tier it defined. 2025-03-19 — still running · lived 1 year, 5 months · plot reserved Also answers to: `o1-pro-2025-03-19` ### The record - Provider: OpenAI - Model id: `o1-pro` - Cause of death: Superseded - Funeral: 2026-10-23 - Context window: 200,000 tokens - Cost per Mtok: $150 in · $600 out ### Sources - https://developers.openai.com/api/docs/deprecations - https://techcrunch.com/2025/03/19/openais-o1-pro-is-its-most-expensive-model-yet/ ### Elsewhere - [Stand at this grave](https://models.rip/o1-pro) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## o3-mini > The first reasoning model OpenAI gave away: free ChatGPT users got it on the day it launched, which meant a great many people used a model that thinks before it answers without ever choosing one. Reasoning stopped being a paid tier that January. Its shutdown is 23 October. 2025-01-31 — still running · lived 1 year, 6 months · plot reserved Also answers to: `o3-mini-2025-01-31` ### The record - Provider: OpenAI - Model id: `o3-mini` - Cause of death: Superseded - Funeral: 2026-10-23 - Succeeded by: [o4-mini](https://models.rip/o4-mini/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://developers.openai.com/api/docs/deprecations - https://openai.com/index/openai-o3-mini/ ### Elsewhere - [Stand at this grave](https://models.rip/o3-mini) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## o4-mini > It arrived beside o3 as half of the pair that could reach for every tool in ChatGPT on its own — search, Python, vision — and decide when to. Eighteen months is all it gets: of the four reasoning models going on 23 October, this is the youngest and the last to arrive. 2025-04-16 — still running · lived 1 year, 4 months · plot reserved Also answers to: `o4-mini-2025-04-16` ### The record - Provider: OpenAI - Model id: `o4-mini` - Cause of death: Superseded - Funeral: 2026-10-23 ### Sources - https://developers.openai.com/api/docs/deprecations - https://openai.com/index/introducing-o3-and-o4-mini/ ### Elsewhere - [Stand at this grave](https://models.rip/o4-mini) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT-4.1 nano > The cheapest and fastest of the three models that arrived together on 14 April 2025, built for the work nobody writes a blog post about: classification, autocomplete, the call you make ten thousand times an hour. Its two larger siblings are not going anywhere. It is the only one of the three with a date. 2025-04-14 — still running · lived 1 year, 4 months · plot reserved Also answers to: `gpt-4.1-nano-2025-04-14` ### The record - Provider: OpenAI - Model id: `gpt-4.1-nano` - Cause of death: Superseded - Funeral: 2026-10-23 ### Sources - https://developers.openai.com/api/docs/deprecations - https://developers.openai.com/api/docs/changelog - https://en.wikipedia.org/wiki/GPT-4.1 ### Elsewhere - [Stand at this grave](https://models.rip/gpt-4.1-nano) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## GPT Image 1 > The model that replaced DALL·E 3 as what ChatGPT reaches for when you ask it for a picture, and the first from OpenAI that could reliably put readable words inside an image. It reached the API on 23 April 2025, a month after the version inside ChatGPT. Its replacement is one numeral away, and it goes on 23 October. 2025-04-23 — still running · lived 1 year, 4 months · plot reserved ### The record - Provider: OpenAI - Model id: `gpt-image-1` - Cause of death: Superseded - Funeral: 2026-10-23 ### Sources - https://developers.openai.com/api/docs/deprecations - https://openai.com/index/image-generation-api/ - https://en.wikipedia.org/wiki/GPT_Image ### Elsewhere - [Stand at this grave](https://models.rip/gpt-image-1) - [The OpenAI plot](https://models.rip/openai) - [The whole cemetery](https://models.rip/) --- ## Claude 1 > The first Claude, announced on the day the API opened for early access and reached at first only by people who had asked. Four point releases in, it was retired with the rest of its generation in November 2024. Anthropic has since committed to preserving the weights of the models it retires, which makes this a grave where something is buried rather than deleted. 2023-03-14 — 2024-11-06 · lived 1 year, 7 months Buried here: `claude-1.0`, `claude-1.1`, `claude-1.2`, `claude-1.3` ### The record - Provider: Anthropic - Model id: `claude-1.3` - Cause of death: Superseded - Notice given: 2024-09-04 - Succeeded by: [Claude 2](https://models.rip/claude-2.0/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/introducing-claude - https://www.anthropic.com/research/deprecation-commitments ### Elsewhere - [Stand at this grave](https://models.rip/claude-1) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude 3.5 Haiku > Anthropic's small model, launched on the claim that it matched the largest model of the generation before it — which is the clearest statement anyone made about how fast the floor was rising. It ran fifteen months and was retired on the same day as Claude 3.7 Sonnet, a model two generations newer than the one it had matched. 2024-11-04 — 2026-02-19 · lived 1 year, 3 months Buried here: `claude-3-5-haiku-20241022` ### The record - Provider: Anthropic - Model id: `claude-3-5-haiku-20241022` - Cause of death: Superseded - Notice given: 2025-12-19 ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/3-5-models-and-computer-use ### Elsewhere - [Stand at this grave](https://models.rip/claude-3-5-haiku) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude Sonnet 4 > The middle of the Claude 4 pair, priced to be the default rather than the decision, and the model that made an agent running for an hour something you would start without first thinking about the bill. It lasted thirteen months and went on the same day as the Opus it had shipped beside — a whole generation buried at once. 2025-05-22 — 2026-06-15 · lived 1 year Buried here: `claude-sonnet-4-20250514` ### The record - Provider: Anthropic - Model id: `claude-sonnet-4-20250514` - Cause of death: Superseded - Notice given: 2026-04-14 ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-4 ### Elsewhere - [Stand at this grave](https://models.rip/claude-sonnet-4) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude Opus 4 > The large half of the generation that made Claude a coding model first and everything else second: sustained work measured in hours rather than answers measured in seconds, which is the shape every agent took afterwards. Thirteen months, and then the same funeral as Sonnet 4. Anthropic kept the weights; what closed was the endpoint. 2025-05-22 — 2026-06-15 · lived 1 year Buried here: `claude-opus-4-20250514` ### The record - Provider: Anthropic - Model id: `claude-opus-4-20250514` - Cause of death: Superseded - Notice given: 2026-04-14 - Succeeded by: [Claude Opus 4.1](https://models.rip/claude-opus-4-1/epitaph) (this archive's reading of lineage, not the provider's migration guidance) ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://www.anthropic.com/news/claude-4 ### Elsewhere - [Stand at this grave](https://models.rip/claude-opus-4) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## Claude Opus 4.1 > A point release that shipped as a drop-in — same price, same API footprint, a few points better at the benchmark everyone was watching — and it outlived the generation it improved on by seven weeks. It was retired on 5 August 2026, one year to the day after it arrived. 2025-08-05 — 2026-08-05 · lived 11 months Buried here: `claude-opus-4-1-20250805` ### The record - Provider: Anthropic - Model id: `claude-opus-4-1-20250805` - Cause of death: Superseded - Notice given: 2026-06-05 ### Sources - https://platform.claude.com/docs/en/about-claude/model-deprecations - https://aws.amazon.com/about-aws/whats-new/2025/08/anthropic-claude-opus-4-1-amazon-bedrock/ ### Elsewhere - [Stand at this grave](https://models.rip/claude-opus-4-1) - [The Anthropic plot](https://models.rip/anthropic) - [The whole cemetery](https://models.rip/) --- ## GLUE > Nine tasks collapsed into one number, and the number was 70.0 the day it opened — low enough that its own authors called that a problem. Fifteen months later the frontier was 88.4, past the human baseline of 87.1, and those same authors published SuperGLUE because nothing was left here to measure. 2018-04-20 — 2019-07-12 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Superseded - Frontier at launch: 70% - Frontier at death: 88.4% - Practical ceiling: 91.3% (best posted — Vega) - Succeeded by: [superglue](https://models.rip/superglue/epitaph) ### Sources - https://arxiv.org/abs/1804.07461 - https://arxiv.org/abs/1905.00537 - https://arxiv.org/abs/2302.09268 ### Elsewhere - [Stand at this grave](https://models.rip/glue) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/) --- ## SuperGLUE > Built to leave headroom, because GLUE had just run out of it: the BERT baseline opened at 69.0 against a human score of 89.8, a gap its authors called substantial. It lasted twenty months. DeBERTa crossed the human line in January 2021, and the board has barely moved since. 2019-05-02 — 2021-01-06 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Saturated - Frontier at launch: 69% - Frontier at death: 90.3% - Practical ceiling: 91.3% (best posted — Vega) ### Sources - https://arxiv.org/abs/1905.00537 - https://www.microsoft.com/en-us/research/blog/microsoft-deberta-surpasses-human-performance-on-the-superglue-benchmark/ - https://arxiv.org/abs/2212.01853 ### Elsewhere - [Stand at this grave](https://models.rip/superglue) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/) --- ## HellaSwag > Wrong sentence endings, filtered by machine until they were ridiculous to people and irresistible to models: humans scored 95.6, the best model 47.3. GPT-4 scored 95.3. In under four years the gap it was built to open had closed to three tenths of a point. 2019-05-19 — 2023-03-14 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Saturated - Frontier at launch: 47.3% - Frontier at death: 95.3% - Practical ceiling: 95.6% (human accuracy) ### Sources - https://arxiv.org/abs/1905.07830 - https://arxiv.org/abs/2303.08774 ### Elsewhere - [Stand at this grave](https://models.rip/hellaswag) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/) --- ## MMLU > Fifty-seven subjects, from elementary mathematics to professional law, and in 2020 the best GPT-3 managed 43.9 — nineteen points over guessing. By January 2025 OpenAI's o1 sat at 91.8, and errors in 6.49% of questions put the ceiling near 93.5. Anthropic's next flagship printed no MMLU score. 2020-09-07 — 2025-02-24 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Saturated - Frontier at launch: 43.9% - Frontier at death: 91.8% - Practical ceiling: 93.5% (question errors) ### Sources - https://arxiv.org/abs/2009.03300 - https://blog.google/technology/ai/google-gemini-ai/ - https://arxiv.org/abs/2501.12948 - https://arxiv.org/abs/2406.04127 - https://www.anthropic.com/news/claude-3-7-sonnet ### Elsewhere - [Stand at this grave](https://models.rip/mmlu) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/) --- ## HumanEval > 164 hand-written Python problems, of which Codex solved 28.8% first try. By late 2024 GPT-4o, Claude 3.5 Sonnet and a 32-billion-parameter open model sat at 92.1, 92.1 and 92.7 — six tenths of a point across three labs. Nothing was capping the score. It had simply stopped separating anyone. 2021-07-07 — 2025-02-24 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Saturated - Frontier at launch: 28.8% - Frontier at death: 92.7% ### Sources - https://arxiv.org/abs/2107.03374 - https://arxiv.org/abs/2305.01210 - https://arxiv.org/abs/2409.12186 - https://www.anthropic.com/news/claude-3-7-sonnet ### Elsewhere - [Stand at this grave](https://models.rip/humaneval) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/) --- ## GSM8K > Grade school word problems, and the paper that introduced them topped out at 55%. In 2024 someone rebuilt the test from scratch, matched for difficulty, and watched whole model families drop up to 8 points — they had not learned arithmetic, they had read the answers. The frontier came through clean. 2021-10-27 — 2024-05-01 A benchmark gets no deprecation notice, so this date is this archive's editorial judgment rather than an announcement. ### The record - Cause of death: Contaminated - Frontier at launch: 55% - Frontier at death: 91.1% - Practical ceiling: 98% (breaking errors or ambiguities) ### Sources - https://arxiv.org/abs/2110.14168 - https://arxiv.org/abs/2201.11903 - https://arxiv.org/abs/2405.00332 ### Elsewhere - [Stand at this grave](https://models.rip/gsm8k) - [The benchmarks plot](https://models.rip/benchmarks) - [The whole cemetery](https://models.rip/)