DeepMind AI Projects List: From AlphaGo to AlphaGenome and Beyond

DeepMind started in 2010 as a London research lab. Now it is one of the biggest AI companies in the world. This DeepMind AI projects list covers their major work.

AlphaGo beat the world champion at Go in 2015. That was the first big win. AlphaFold solved the protein folding problem in 2020. It won the 2024 Nobel Prize for that work. AlphaGenome cracked 98% of gene dark matter in 2026. AlphaCode writes computer programs. AlphaEvolve designs algorithms that improve Google's chips.

Gemini is DeepMind's family of large language models. Gemini 3.5 handles complex tasks for companies like Shopify and Salesforce. Co-Scientist now runs physical lab experiments. It grew semiconductors in about an hour. GraphCast predicts weather. Gemini Robotics controls robots.

From games to science to healthcare, DeepMind is building AI that matters. This article covers the full list.

The Alpha Series: DeepMind's Most Famous Projects

AlphaGo

AlphaGo was the project that put DeepMind on the map. In 2015, it became the first AI program to defeat a world champion Go player . Go is an ancient Chinese board game far more complex than chess. Many experts thought AI would not master it for another decade.

In 2017, DeepMind released AlphaGo Zero. It learned entirely by playing against itself, with no human data. It defeated the original AlphaGo 100-0 . AlphaZero then generalized this approach to master chess, shogi, and Go with one algorithm .

AlphaFold

AlphaFold is perhaps DeepMind's most important scientific contribution. Launched in 2020, it solved the 50-year grand challenge of predicting protein structures from amino acid sequences . This has enormous implications for biology and medicine.

AlphaFold 3, released in May 2024, can predict structures and interactions of almost all biological molecules including proteins, DNA, RNA, and small molecules . In 2026, it achieved 98.5% accuracy on human protein structures . AlphaFold was awarded the 2024 Nobel Prize in Chemistry .

AlphaStar

AlphaStar reached Grandmaster level in StarCraft II in 2019 . This game requires real-time decision-making and handling imperfect information. It showed that DeepMind's AI could handle complex, fast-paced environments.

AlphaCode

In 2022, DeepMind released AlphaCode, a system that generates computer programs . It can solve competitive programming problems at a level comparable to human programmers.

AlphaGenome

Announced in January 2026, AlphaGenome is a gene decoding tool that has cracked 98% of "gene dark matter" with 90% accuracy . This helps scientists understand previously unknown functions in human DNA.

AlphaEvolve

AlphaEvolve is a Gemini-powered coding agent that designs advanced algorithms . It has been used to improve DNA sequencing error correction by 30%, optimize electricity grids, and even design components for Google's TPU chips . In financial services, Klarna used it to double training speed for a transformer model while improving quality .

AlphaProof

AlphaProof showed DeepMind's strength in reasoning by winning medals in mathematical Olympiads .

Read More: DeepMind AI Software Features: A Complete Guide to Googles AI Models

Gemini and Large Language Models

DeepMind has also built the Gemini series of AI models. Gemini 3.5 is described as offering "frontier intelligence with action" . The models handle agentic coding, advanced multimodal understanding, and long-horizon tasks .

The Gemini models are being used by companies like Shopify, Salesforce, Macquarie Bank, and Databricks for complex enterprise tasks . For example, Macquarie Bank uses Gemini 3.5 to accelerate customer onboarding by reasoning over 100+ page documents .

Gemini Omni is a model that creates anything from anything, starting with video. Gemini Image (Nano Banana) is a state-of-the-art image generation and editing model built on Gemini .

Games Research: The Foundation of DeepMind's Breakthroughs

Games have driven many of DeepMind's biggest breakthroughs . From Atari to Go to StarCraft, games provide a controlled environment for testing AI.

Project SIMA

SIMA (Scalable Instructable Multiworld Agent) is a generalist agent that "sees" what a player would see on screen, understands natural language instructions, and acts through keyboard and mouse controls . Powered by Gemini, SIMA 2 acts as an interactive companion capable of real-time reasoning and conversation. It has been tested in games including No Man's Sky, Valheim, and Hydroneer .

EVE Online Partnership

DeepMind has partnered with Fenris Creations to research AI within the EVE Online universe . EVE Online's persistent world, real economy, and complex player dynamics provide unique challenges for AI research, including continual learning, long-horizon planning, and multi-agent dynamics .

AI for Science: Co-Scientist and Laboratory Automation

Co-Scientist

Co-Scientist is a multi-agent AI system built on Gemini that accelerates scientific research . It uses specialized agents to generate ideas, debate hypotheses, and evolve solutions through a "tournament of ideas" approach .

In validation studies, Co-Scientist helped uncover drug repurposing candidates for liver fibrosis, with one candidate blocking 91% of a scarring-linked response in lab tests . It has also helped researchers on ALS, reversing cellular aging, and metabolic liver disease .

In September 2026, Google DeepMind and Duke University released a major upgrade. Co-Scientist can now perform physical experiments in the lab . It generated a synthesis recipe for MXene materials and successfully grew high-quality crystals on the first try. It also synthesized high-quality single-layer MoS₂, MoSe₂, and WS₂ semiconductors - materials crucial for next-generation electronics - in about an hour . This is the first time an AI system has directly controlled a semiconductor manufacturing device.

World Models: Genie 1, 2, and 3

DeepMind has been building world models that can create interactive environments. Genie 1 generated 2D platformers from images. Genie 2 generates "infinite" playable 3D worlds from single images and text descriptions . Genie 3 creates high-quality, photorealistic 3D worlds that can be interacted with in real time .

AI for Weather and Climate

GraphCast

GraphCast is a spherical graph neural network that provides accurate 15-day weather forecasts . It moves away from traditional physics simulations toward AI-driven prediction.

GenCast

GenCast is a probabilistic weather model that is more efficient and accurate than gold-standard benchmarks 97% of the time .

FGN (Functional Generative Network)

FGN directly predicts cyclone behavior and is now being used by the US National Hurricane Center .

Robotics and Physical AI

Gemini Robotics

DeepMind launched Gemini Robotics in March 2025. It is a vision-language-action model that helps robots understand new situations without specialized training . A second model, Gemini Robotics-ER (Embodied Reasoning), is a vision-language model that "understands complex and dynamic worlds" .

In July 2026, DeepMind released Gemini Robotics ER 2, a major upgrade .

Newton Platform

DeepMind collaborated with NVIDIA and Disney Research to develop the Newton physics engine for robotics .

D4RT

In January 2026, DeepMind released D4RT (Dynamic 4D Reconstruction and Tracking) .

AI for Health and Bioresilience

DeepMind and its spin-off Isomorphic Labs are also working on bioresilience. Their work focuses on prevention, detection, and response to biological threats .

AlphaFold and AlphaGenome are being used to design vaccines and medical countermeasures. Isomorphic Labs has established a unit to rapidly deploy its drug design engine during novel outbreaks .

You May Also Read: How DeepMind Improves Google Search AI in 2026: Top Updates

Other Notable Projects

Sparrow

Sparrow was an AI chatbot released in 2022 as a research preview .

Pi

Pi was a 2023 AI "friend" designed for conversation rather than as a productivity tool .

V2A (Video-to-Audio)

V2A generates synchronized soundtracks from video content and prompts .

SIMI

SIMI (Scalable Instructable Multiworld Agent) was released in 2024 .

Zipper

Zipper is a 2024 architecture that combines multiple single-modal pre-trained decoder models .

AlphaChip

AlphaChip uses reinforcement learning to optimize chip component layout, significantly reducing design time. It is already used in Google's TPUs .

How to Access DeepMind Projects?

• AlphaFold Server is available online with usage limits for academic researchers .
• Gemini models are available through Google AI Studio, Gemini API, and Gemini Enterprise Agent Platform .
• Co-Scientist is being previewed with organizations including Bayer, Daiichi Sankyo, and the US National Laboratories .

The Bottom Line

DeepMind has moved from game-playing AI to solving real-world problems in science, healthcare, energy, and manufacturing. The DeepMind AI projects list shows a clear trajectory. From AlphaGo's victory in 2015 to AlphaFold's Nobel Prize in 2024 and Co-Scientist's automated lab experiments in 2026, DeepMind is consistently pushing boundaries.

Its AI systems are now designing computer chips, predicting protein structures, controlling semiconductor manufacturing, and helping prepare for future pandemics. This is AI that actually matters.

FAQs

1. What is DeepMind best known for?

AlphaFold. It solved the protein folding problem in 2020. That was a 50-year old scientific challenge. It won the 2024 Nobel Prize in Chemistry. AlphaGo is also famous for beating the world champion at Go in 2015.

2. What are the main DeepMind AI projects?

AlphaGo, AlphaFold, AlphaGenome, AlphaCode, AlphaEvolve, and AlphaProof. Gemini is their family of large language models. Co-Scientist does automated lab experiments. GraphCast predicts weather. Gemini Robotics controls physical robots.

3. What is AlphaGenome?

It is a gene decoding tool released in January 2026. It cracked 98% of gene dark matter with 90% accuracy. This helps scientists understand previously unknown functions in human DNA.

4. What does Co-Scientist do?

It is a multi-agent AI system that accelerates scientific research. It generates ideas and tests hypotheses. In September 2026, it started running physical lab experiments. It grew high-quality semiconductors in about an hour.

5. Can I use DeepMind's AI tools?

Yes. AlphaFold Server is available online for academic researchers. Gemini models are available through Google AI Studio and the Gemini API. Co-Scientist is being previewed with organizations like Bayer and the US National Laboratories.