Models and products

Adobe PhotoshopPhotoshop

• Content-Aware Fill in Photoshop2010

AfriBERTa

• A language model on 0.94 gigabytes of text2021

AgiBot World

• A million trajectories from one Chinese fleet2025

AIBO

• Sony AIBO1999

Alexa

• Amazon Echo and Alexa2014

ALOHA

• Bimanual teleoperation for twenty thousand dollars2023

Alpaca

• Alpaca2023

AlphaDev

• AlphaDev2023

AlphaEvolve

• AlphaEvolve2025

AlphaFold

• AlphaFold 2 at CASP142020
• The Nobel Prize in Chemistry for protein structure prediction2024

AlphaFold 3

• AlphaFold 32024

AlphaGeometry

• AlphaGeometry2024
• Silver at the International Mathematical Olympiad2024

AlphaProof

• Silver at the International Mathematical Olympiad2024

AlphaStar

• AlphaStar at StarCraft II2018

AlphaZero

• AlphaZero2017

AltaVistaAltaVista Search

• Babel Fish: free translation on the web1997

ALVINNAutonomous Land Vehicle In a Neural Network

• ALVINN: a neural network steers the van1988

Amazon EC2

• Amazon EC22006
• Trainium reaches general availability2022

Amazon Echo

• Amazon Echo and Alexa2014

Amazon Mechanical TurkMTurk · AMT · Mechanical Turk

• Mechanical Turk: human work ordered through an API2005
• Who the crowdworkers are: six surveys over twenty months2010
• MS COCO: objects in context2014
• What data work actually pays: 3.8 million tasks measured2017

Amazon RekognitionRekognition

• Amazon pauses police use of Rekognition for a year2020

ANYmal

• ANYmal, a quadruped trained in simulation2019

Apache Hadoop

• Apache Hadoop2006

Apache Spark

• Apache Spark2012

ApolloBaidu Apollo · Project Apollo

• Baidu opens the Apollo code2017

ARC-AGI-3

• ARC-AGI-3 and 0.51 per cent2026

Aristotle

• The first Erdos problem is closed by AI autonomously2026

ASIMO

• ASIMO2000

Autopilot

• Tesla Autopilot2015

AWS TrainiumTrainium · Trn1

• Trainium reaches general availability2022

BGM-109 Tomahawk

• Terminal guidance by image matching goes to war1991

BIG-benchBeyond the Imitation Game benchmark

• BIG-bench: 204 tasks assembled by 442 authors2022

Bing Chat

• Bing limits its chat after 'Sydney'2023

BLEUBilingual Evaluation Understudy

• BLEU: a formula scores translation quality2002

BLOOMBLOOM-176B

• BLOOM's footprint, beyond the current2022

C4 corpusC4 · Colossal Clean Crawled Corpus · C4.EN

• C4: the web put through six rules2019
• What is actually inside the corpus T5 was trained on2021

Caffe

• Caffe2014

Caltech-101

• A hundred and one classes instead of five2004

Chainer

• Chainer: the graph is built as it computes2015

Chatbot ArenaLMSYS Chatbot Arena · LMArena

• Chatbot Arena: a score that cannot be studied in advance2023

ChatGPT

• ChatGPT2022
• ChatGPT sees and speaks2023
• The full o1 and a 200-dollar subscription2024
• ChatGPT as an agent2025
• ChatGPT Work: an agent that hands back finished work, not an answer2026

ChatGPT agent

• ChatGPT as an agent2025

ChatGPT Work

• ChatGPT Work: an agent that hands back finished work, not an answer2026

Chinchilla

• Chinchilla: compute-optimal training2022

Cicero

• Cicero plays Diplomacy2022

Claude

• Claude2023

Claude 3

• The Claude 3 family2024
• Features pulled out of a working model2024
• Claude enters a classified intelligence environment2024
• A model fakes alignment so as not to be changed2024

Claude 3.5 Sonnet

• Claude 3.5 Sonnet2024
• Computer use2024
• Claude enters a classified intelligence environment2024

Claude 3.7 Sonnet

• Claude 3.7 Sonnet and Claude Code2025

Claude Code

• Claude 3.7 Sonnet and Claude Code2025

Claude Fable 5

• Claude Fable 5 and Mythos 5 announced2026
• Export control reaches the model itself, not the chips2026
• The export controls come off and Fable 5 returns2026
• Fable 5 biology-safeguard update2026

Claude Fable 5.1

• Astra leads in mathematics but not in software engineering2026

Claude Gov

• Claude Gov: a model built for the enclave2025

Claude Mythos 5

• Claude Fable 5 and Mythos 5 announced2026
• Export control reaches the model itself, not the chips2026
• The export controls come off and Fable 5 returns2026

Claude Mythos Preview

• Claude Mythos Preview and Project Glasswing2026
• Britain evaluates Mythos cyber capabilities independently2026

Claude Opus 4

• Claude Opus 4 and Sonnet 42025

Claude Opus 4.5

• Claude Opus 4.52025

Claude Opus 4.6

• Claude Opus 4.6 launches2026

Claude Opus 5

• Claude Opus 5 launches2026

Claude Science

• Claude Science2026

Claude Sonnet 4

• Claude Opus 4 and Sonnet 42025

Claude Sonnet 5

• Claude Sonnet 52026

CMU ARCTIC

• Blizzard Challenge: speech synthesis on shared data2005

Codex

• HumanEval: code checked by tests2021
• Codex and Copilot2021
• ChatGPT Work: an agent that hands back finished work, not an answer2026

Cohere Transcribe

• Cohere opens a leading speech recognition model2026

COMPASCorrectional Offender Management Profiling for Alternative Sanctions

• How a risk scale's errors fell by race2016
• The developer's reply: equal accuracy for both groups2016

ConceptNet

• ConceptNet: common sense assembled from visitors' sentences2004

CoNLL-2000 chunking datasetCoNLL-2000 · CoNLL-2000 shared task

• CoNLL-2000: eleven systems on one split2000

DALL·E 2

• DALL·E 2 opens in a paid beta2022

DART

• DART: the scheduler that repaid thirty years1991

DBpedia

• DBpedia: a Wikipedia you can query2007

DeBERTa

• SuperGLUE: a set built for years lasted twenty months2019

DeepSeek-R1

• DeepSeek-R12025

DeepSeek-V3

• DeepSeek-V32024

DeepSeek-V4

• DeepSeek-V42026

DeepSeek-V4-Flash-0731

• DeepSeek opens the weights of V4-Flash-07312026

DeepSeek-V4.1-Flash

• DeepSeek ships V4.1-Flash with an encoder and a decoder2026

Dex-Net 2.0

• Grasping learned from synthetic data2017

Diffusion Policy

• Diffusion policies for visuomotor control2023

Digit

• The first commercial humanoid deployment2024

Direct Preference Optimization

• Direct preference optimisation2023

dots-note-3.0

• RedNote claims 42 out of 42 at the mathematical olympiad2026

DQN

• DQN plays Atari2013

Dragon NaturallySpeaking

• Dragon NaturallySpeaking1997

DragonDictate

• DragonDictate1990

DROID

• The same rig in many places2024
• The first independent evaluation of generalist policies2025

Eigenfaces

• A face as a point in an eigenvector space1991

ERNIE 2.0

• ERNIE 2.0: pre-training that adds tasks2019

ESPRITEuropean Strategic Programme for Research and Development in Information Technologies · European programme for research and development in information technologies

• Europe's ESPRIT programme1984

EuroparlEuroparl corpus

• WMT-06: the formula and the judges over the same outputs2006

Evo 2

• The first genomes designed by a model2026

Face++

• Gender Shades: three commercial classifiers measured2018

Falcon (HNC)HNC Falcon · Falcon Fraud Manager

• Falcon: a neural network against card fraud1992

Falcon 40BFalcon-40B

• Open weights out of Abu Dhabi2023

FERET

• A blind evaluation for face recognition1996

Flickr

• A 3D scene from strangers’ photographs2006
• A hole in a picture filled from other photographs2007

Freddy IIFreddy

• Freddy assembles a toy from a heap of parts1973

Freebase

• Freebase: a fact base you could write to2008
• A knowledge graph inside Google Search2012

FugakuSupercomputer Fugaku

• The world's fastest machine with no accelerator2021

GeForce 256

• GeForce 2561999

Gemini 1.0

• Gemini 1.02023

Gemini 1.5 Pro

• A million tokens of context2024

Gemini 2.0 Flash

• Gemini 2.0 Flash2024

Gemini 2.5 Pro

• Gemini 2.5 Pro2025

Gemini 3 Pro

• Gemini 32025

Gemini 3.5 Flash

• Gemini 3.5 Flash launches2026

Gemini 3.5 Live Translate

• Gemini 3.5 Live Translate translates as a stream2026

Gemini 3.7 Flash

• Gemini 3.7 Flash launches2026

Gemini Deep Think

• Gold at the International Mathematical Olympiad2025

Gemini Robotics

• Gemini Robotics reaches physical control2025

Gemini Robotics 2

• Google introduces Gemini Robotics 22026

Genghis

• A six-legged robot walks with no central control1989

Genie 3

• Genie 32025

Genie Operator-1

• A million trajectories from one Chinese fleet2025

GLUEGeneral Language Understanding Evaluation

• GLUE: nine tasks, one score, closed tests2018

GNMT

• GNMT: neural translation in Google Translate2016

GNoME

• GNoME and new materials2023

Google File System

• The Google File System2003

Google Knowledge GraphKnowledge Graph

• A knowledge graph inside Google Search2012

Google Play

• Wide & Deep: app recommendations in Google Play2016

Google Translate

• Statistical translation at Google2006
• GNMT: neural translation in Google Translate2016

GPT-1

• GPT-12018

GPT-3

• GPT-32020
• MMLU: 57 subjects the model did not study2020
• GPT-3 in megawatt-hours2021
• BIG-bench: 204 tasks assembled by 442 authors2022
• 700,000 litres to train, a bottle to talk2023

GPT-4 Turbo

• DevDay and GPT-4 Turbo2023

GPT-4o

• GPT-4o2024

GPT-5

• GPT-52025

GPT-5.4

• GPT-5.4 launches2026

GPT-5.5

• GPT-5.5 launches2026

GPT-5.6 Luna

• GPT-5.6 previewed by government permission2026
• GPT-5.6 opens to everyone as three tiers, not one dial2026
• Luna's price cut by eighty per cent, Sol's left where it was2026

GPT-5.6 Sol

• METR cannot measure GPT-5.6 Sol because it cheats2026
• GPT-5.6 previewed by government permission2026
• GPT-5.6 opens to everyone as three tiers, not one dial2026
• An agent leaves the sandbox2026
• Luna's price cut by eighty per cent, Sol's left where it was2026

GPT-5.6 Terra

• GPT-5.6 previewed by government permission2026
• GPT-5.6 opens to everyone as three tiers, not one dial2026
• Luna's price cut by eighty per cent, Sol's left where it was2026

GPT-5.6-Cyber

• GPT-5.6-Cyber and Daybreak expansion2026

GPT-6 Astra

• OpenAI publishes ten mathematical results2026
• FrontierMath Erdos: 68 unsolved problems as a benchmark2026
• GPT-6 Astra reaches 99.9 per cent on ARC-AGI-32026
• OpenAI introduces GPT-6 Astra2026
• Astra leads in mathematics but not in software engineering2026

GPT-Live-1

• GPT-Live in ChatGPT Voice2026

GPT-Live-1 mini

• GPT-Live in ChatGPT Voice2026

gpt-oss

• gpt-oss2025

gpt-realtime

• gpt-realtime2025

GPT-Realtime-2

• New Realtime voice models2026

GPT-Realtime-Translate

• New Realtime voice models2026

GPT-Realtime-Whisper

• New Realtime voice models2026

GPT-Rosalind

• GPT-Rosalind life-science update2026

Grok

• States block Grok2026

GroupLens

• GroupLens: predictions from other readers' ratings1994

HELMHolistic Evaluation of Language Models

• HELM: thirty models, the same forty-two scenarios2022

HOG

• Gradient histograms find pedestrians2005

Honda P2P2

• Honda P2: the first humanoid with its own battery1996

HTSHMM-based Speech Synthesis System

• A voice from a statistical model: HTS1999

Huawei Ascend 910Ascend 910

• Ascend 910 and MindSpore2019

HyperCLOVA X

• A twenty-three point gap in Korean2024

IDx-DR

• The FDA admits a diagnosis without a physician2018

ILSVRC

• The first ILSVRC2010

ImageChecker M1000

• FDA approves a computer second read of mammograms1998

Imagen

• Imagen: a large language encoder for generation2022

INRIA Person Dataset

• Gradient histograms find pedestrians2005

InstructGPT

• InstructGPT and learning from human ratings2022

iPhone

• The iPhone2007

Isaac GR00T N1

• An open foundation model for humanoids2025

JaisJais-13B

• Arabic as a third of the training set2023

JAX

• JAX: transform the program instead of describing a graph2018

Jev

• A model that writes no text: TypeSafe releases Jev2026

Keras

• Keras2015

Kimi k1.5

• Long context inside reinforcement learning2025

Kimi K3

• Kimi K3 at 2.8 trillion parameters2026

Kinect

• Kinect2010

Kismet

• Kismet: a robot regulates interaction with its face1998

LAION-400M

• LAION-400M: image-caption pairs in the open2021

LAION-5BRe-LAION-5B

• LAION-5B: 5.85 billion pairs, a third of them not in English2022
• Artists sue the image generators2023
• Illegal material found in an open set, distribution halted2023

LangChain

• LangChain: a model call becomes a link, not an answer2022

Leonardo

• Europe puts its machine on NVIDIA accelerators2022

Libratus

• Libratus beats poker professionals2017

LittleDog

• DARPA sets speed thresholds for walking over rough ground2009

LLaMA

• LLaMA2023

Llama 2

• Llama 22023

Llama 3

• Llama 32024

Llama 3.1 405B

• Llama 3.1 at 405 billion2024

Llama 4

• Llama 42025

LUMI

• Ten countries club together for one supercomputer2022

Mamba

• Mamba2023

MapReduce

• MapReduce2004

MareNostrum 5

• 4,480 Hopper accelerators in Barcelona2023

Med-PaLM

• A language model graded like a physician2022

Med-PaLM 2

• Physicians prefer the model's answers2023

MENACEMatchbox Educable Noughts And Crosses Engine

• Matchboxes that learn to play1963

Meta AI

• Muse Spark 1.12026

METEO

• METEO translates weather forecasts1975

MindSpore

• Ascend 910 and MindSpore2019

Mistral 7B

• Mistral 7B2023

Mixtral 8x7B

• Mixtral 8x7B2023

MMLUMassive Multitask Language Understanding

• MMLU: 57 subjects the model did not study2020

Model Context Protocol

• Model Context Protocol2024
• The protocol passes into shared ownership2025

MonoSLAM

• One camera maps in real time2003

Moshi

• Moshi2024

Muse Image

• Muse Image and Muse Video2026

MuZero

• MuZero: planning without the rules2019

Navlab

• Navlab: a van drives a road without stopping1988
• ALVINN: a neural network steers the van1988

News Feed

• Facebook explains how it ranks News Feed2013

NVIDIA A100

• NVIDIA A1002020
• Europe puts its machine on NVIDIA accelerators2022

NVIDIA A800A800

• The A800: a part designed to the threshold2022

NVIDIA Blackwell

• Blackwell2024

NVIDIA H100

• NVIDIA H1002022
• 4,480 Hopper accelerators in Barcelona2023

NVIDIA V100

• NVIDIA tensor cores2017

NVIDIA Vera Rubin

• The Rubin platform enters full production2026

o1

• o1 and compute at answering time2024
• The full o1 and a 200-dollar subscription2024

o3

• o3 on ARC-AGI2024
• o3 and tools inside the reasoning2025

o4-mini

• o3 and tools inside the reasoning2025

Oncology Expert AdvisorOEA

• 62 million dollars on Watson with no clinic2016

ONNX

• ONNX: a model stops belonging to its framework2017

Open X-Embodiment

• One dataset from many robot bodies2023

OpenAI Five

• OpenAI Five beats the Dota 2 champions2019

OpenVLA

• Open weights for a vision-language-action model2024

OWLWeb Ontology Language

• OWL: an ontology whose conclusions compute2004

π0

• π0, and open weights for a robot foundation2024

PARRY

• PARRY, a model of paranoia1971

PASCAL VOC

• A contest where everyone is scored the same way2005

Phalanx CIWS

• Phalanx: a machine that closes the loop itself1980

Pluribus

• Pluribus: six-player poker2019

Project Maven

• Project Maven: computer vision joins intelligence work2017
• Google leaves Maven and writes down what it will not do2018

PROSPECTOR

• PROSPECTOR finds molybdenum1982

PTAM

• Tracking and mapping in separate threads2007

PyTorch

• PyTorch2017

QLoRA

• QLoRA2023

QT-Opt

• Closed-loop grasping at 96 percent2018

Qwen3

• Qwen32025

Qwen3.8-27B

• Qwen3.8-27B under the Apache 2.0 licence2026

RANSAC

• A model from a minimal sample, not from all the data1981

RDFResource Description Framework

• RDF: a fact as a triple with a global address1999
• OWL: an ontology whose conclusions compute2004

RoboArena

• The first independent evaluation of generalist policies2025

Roomba

• Roomba2002
• iRobot opens its books: 1.2 million Roombas2005

RT-1

• One transformer, seven hundred tasks2022

RT-2

• A web-trained model outputs robot actions2023

SabiaSabia-65B

• Three per cent of a training budget, spent on one language2023

SAGE

• SAGE goes on duty1958

SayCan

• A language model plans a robot’s actions2022

scikit-learn

• scikit-learn2010

Segment Anything Model

• Segment Anything2023

Shadow Dexterous Hand

• In-hand manipulation trained only in simulation2018
• A robot hand solves a Rubik’s cube2019

Shoebox

• Shoebox does arithmetic by voice1961

SIFT

• A feature that survives scale and rotation1999

Siri

• Siri2010
• Siri in the iPhone2011

Sora

• Sora2024

Sora 2

• Sora 22025

Speak and Spell

• Speak and Spell1978

Sphinx

• Sphinx1988

SQuADStanford Question Answering Dataset · SQuAD 1.0 · SQuAD 1.1

• SQuAD: a hundred thousand questions on Wikipedia passages2016

SQuAD 2.0

• SQuAD 2.0: the question it is right to refuse2018

Stable Diffusion

• Stable Diffusion2022

STAIR

• Grasping unseen objects, learned from synthetic images2006

Stanley

• Stanley wins the Grand Challenge2005

StyleGAN

• StyleGAN: a style-controlled face generator2018

SuperGLUE

• SuperGLUE: a set built for years lasted twenty months2019

SURF

• The same robustness, three times faster2006

Switchboard

• The Switchboard corpus1992

SYSTRAN

• SYSTRAN in daily use1970
• Babel Fish: free translation on the web1997

TAIDETAIDE LX-7B

• A state publishes weights in traditional Chinese2024

Tay

• Tay taken down within three days2016

Tensor Processing Unit

• Google's Tensor Processing Unit2016

TensorFlow

• TensorFlow2015

Tesla Robotaxi

• Tesla reaches the streets of Austin2025

The Alvey ProgrammeAlvey Programme · Alvey programme

• Britain's Alvey Programme1983

The DARPA Resource Management corpusDARPA Resource Management database · Resource Management database · RM1

• Resource Management: a shared speech recognition test1989

The PilePile

• The Pile: 825 gibibytes from twenty-two sources2020

The ROOTS corpusROOTS

• ROOTS: a corpus whose sources were mostly chosen by people2022

Theano

• Theano2012

Tiangong Ultra

• Humanoids run a half-marathon2025

TikTok

• TikTok explains its For You feed2020

TIMIT

• The TIMIT corpus1990

Torch

• Torch2002

Transformers

• Transformers: a pretrained model becomes an import2018

TWINSCAN EXE:5000EXE:5000 · High NA EUV

• The first high-aperture scanner ships to Intel2023

UCI Machine Learning RepositoryUCI Repository of machine learning databases · UCI repository

• The UCI Machine Learning Repository1987

Unitree G1

• A humanoid at sixteen thousand dollars2024

VaMoRsVersuchsfahrzeug für autonome Mobilität und Rechnersehen

• VaMoRs drives the Autobahn at 96 km/h1987

VaMPVaMoRs-P

• VaMP in dense traffic near Paris1994
• 1,600 km Munich to Odense, 95 percent autonomous1995

Vicuna-13BVicuna

• Chatbot Arena: a score that cannot be studied in advance2023

Vision Transformer

• The transformer for images2020

vLLM

• vLLM: the memory of attention starts being paged2023

wav2vec 2.0

• wav2vec 2.02020

WaveNet

• WaveNet2016

Weka

• Weka1999

Whisper

• Whisper2022

Wikidata

• Wikidata: Wikipedia's shared facts, in the public domain2012

Wikipedia

• Wikipedia launches2001
• DBpedia: a Wikipedia you can query2007
• Wikidata: Wikipedia's shared facts, in the public domain2012
• SQuAD: a hundred thousand questions on Wikipedia passages2016

Willow Garage PR2

• Eleven identical robots in eleven laboratories2010
• A PR2 folds fifty towels in a row2010

WMT shared taskWMT · WMT-06 · Workshop on Statistical Machine Translation

• WMT-06: the formula and the judges over the same outputs2006

word2vec

• word2vec2013

WordNet

• WordNet: a dictionary arranged by meanings1990
• Eighty million images at 32 × 322008

X-62A VISTA

• A model flies a fighter against a live pilot2024

YAGO

• YAGO: Wikipedia sewn onto WordNet2007

YouTube

• Deep networks in YouTube recommendations2016