Google DeepMind

DeepMind was founded in 2010 by neuroscientist Demis Hassabis, Shane Legg, and Mustafa Suleyman, taking an interdisciplinary approach to building general AI systems that combines fields such as machine learning, neuroscience, engineering, and mathematics. Jaan Tallinn, a developer of Skype and Kazaa, was an early investor.⁠[1][1][2][2]

In its early years, the lab became known for pioneering deep reinforcement learning, a combination of deep learning and reinforcement learning, and for using games to test its systems. Its early result DQN, which observed only the raw pixels on the screen and used the game score as its reward, achieved human-level performance in almost half of the 50 Atari 2600 games it was applied to.⁠[1][3][3]

On January 26, 2014, Google announced the acquisition of London-based DeepMind; DeepMind confirmed the deal but did not disclose its terms. The tech publication The Information estimated the price at more than $500 million and reported that Facebook had also held talks to buy the company. At the time, DeepMind had about 75 employees, and Founders Fund and Horizons Ventures were its major investors; Google agreed to establish an ethics board to ensure that its artificial intelligence technology would not be abused.⁠[2][2][2][2][2][2]

AlphaGo combines deep neural networks with search algorithms, using a “policy network” to select moves and a “value network” to predict the winner; it first learned from human games and then improved through reinforcement learning by playing against itself. In October 2015, it defeated three-time European champion Fan Hui 5–0, becoming the first program to beat a professional Go player in a formal match; in March 2016, in Seoul, South Korea, it defeated Lee Sedol, winner of 18 world titles, 4–1 in a match watched by over 200 million people worldwide, earning it a professional 9-dan ranking. Its “Move 37” in game two and Lee Sedol’s “Move 78” in game four have since been studied by players of all levels.⁠[4][4][4][4][4][4][4][4][4][4][4][4]

On September 8, 2016, DeepMind released WaveNet, a generative model that directly models raw audio waveforms sample by sample; its synthesized speech sounded more natural than the best text-to-speech systems of the time, reducing the gap with human performance by over 50%, and it could also generate music. WaveNet was later used as the voice of the Google Assistant.⁠[5][5][5][1]

On July 3, 2017, the UK Information Commissioner’s Office ruled that London’s Royal Free hospital had failed to comply with the Data Protection Act when it handed over the personal data of about 1.6 million patients to DeepMind as part of their partnership to develop Streams, an alert app for acute kidney injury. The ruling was largely based on the fact that the app was still being tested after the data was transferred, and patients had not been adequately informed that their data would be used for the test. The ruling did not directly criticize DeepMind, because the ICO regarded the hospital as the data controller and DeepMind as a data processor acting on its behalf. DeepMind responded that when it began the work in 2015, it had underestimated the complexity of the NHS and of the rules around patient data, saying “We got that wrong.”⁠[6][6][6][6][6][6][6][6]

In late 2017, DeepMind introduced AlphaZero, a single system that taught itself chess, shogi, and Go from scratch using only the basic rules, beating a world-champion program in each of the three games; on December 6, 2018, its full evaluation was published in the journal Science. AlphaZero first outperformed the chess engine Stockfish after 4 hours of training, and outperformed the version of AlphaGo that beat Lee Sedol in 2016 after 30 hours of training.⁠[7][7][7][7][7]

On October 30, 2019, DeepMind announced that AlphaStar, playing under the same viewing and action-rate constraints as humans on Battle.net, the official server for the real-time strategy game StarCraft II, was ranked above 99.8% of active players and achieved Grandmaster level with all three races: Protoss, Terran, and Zerg.⁠[8][8]

On November 30, 2020, in the results of the 14th biennial Critical Assessment of protein Structure Prediction (CASP), the new version of AlphaFold achieved a median Global Distance Test (GDT) score of 92.4 across all targets, with an average error of about 1.6 angstroms, and was recognized by the CASP organizers as a solution to the “protein folding problem” that had challenged biology for about 50 years. According to CASP founder John Moult, a score of around 90 is considered competitive with experimental methods.⁠[9][9][9][9]

In 2021, DeepMind open-sourced the AlphaFold code and, in partnership with the European Bioinformatics Institute of the European Molecular Biology Laboratory (EMBL-EBI), launched the free and open AlphaFold Protein Structure Database. On July 28, 2022, the two released predicted structures for nearly all catalogued proteins, expanding the database from nearly 1 million structures to over 200 million, covering plants, bacteria, animals, and other organisms.⁠[10][10][10][10]

On April 20, 2023, Sundar Pichai, CEO of Google and Alphabet, announced that DeepMind and the Brain team from Google Research would be combined into a single unit, Google DeepMind, with Hassabis as CEO, responsible for developing Google’s most capable general AI systems; Jeff Dean took on the role of Google’s Chief Scientist. In a letter to staff, Hassabis announced the creation of a Scientific Board led by Koray Kavukcuoglu, with Eli Collins as VP of Product and Zoubin Ghahramani joining the research leadership team.⁠[11][11][11][12][12]

The Brain team started in 2011 at Google’s X lab, developed open source software such as TensorFlow and JAX as well as sequence-to-sequence learning, and in 2017 invented the Transformer architecture that underpins almost all large language models.⁠[1][1][1]

On December 6, 2023, Google launched Gemini 1.0, introduced in a post by Hassabis on behalf of the Gemini team; the model was developed collaboratively by teams across Google, and Pichai called it the first realization of the vision behind the formation of Google DeepMind. Gemini was designed to be natively multimodal, pre-trained from the start on data in multiple modalities, and came in three sizes: Ultra, Pro, and Nano. Gemini Ultra exceeded state-of-the-art results on 30 of 32 widely used academic benchmarks and, with a score of 90.0% on MMLU, became the first model to outperform human experts on it. On the same day, Bard switched to a fine-tuned version of Gemini Pro.⁠[13][13][13][13][13][13][13][13][13][13]

On March 19, 2024, Mustafa Suleyman, co-founder of DeepMind and Inflection, joined Microsoft as EVP and CEO of the newly formed Microsoft AI, responsible for consumer AI products such as Copilot and related research.⁠[14][14][14]

On May 8, 2024, Google DeepMind and Isomorphic Labs published AlphaFold 3 in Nature. It can predict the structures and interactions of life’s molecules, including proteins, DNA, RNA, and ligands, with at least a 50% improvement over existing methods in predicting the interactions of proteins with other molecules; scientists can use most of its capabilities for free through the newly launched AlphaFold Server. In November 2024, the model code and weights were released for academic use.⁠[15][15][15][15][15]

On October 9, 2024, the Royal Swedish Academy of Sciences announced that one half of the 2024 Nobel Prize in Chemistry would go to David Baker for computational protein design, and the other half jointly to Demis Hassabis and John Jumper of Google DeepMind for protein structure prediction. AlphaFold2, which the two presented in 2020, can predict the structures of virtually all of the roughly 200 million proteins researchers have identified and has been used by more than two million people in 190 countries. At the time of the award, Hassabis was CEO of Google DeepMind and Jumper was a Senior Research Scientist, with both affiliated with Google DeepMind in London.⁠[16][16][16][16][16][16][16][17]

On December 11, 2024, Google launched Gemini 2.0, aimed at the “agentic era,” adding native image and audio output and native tool use; the experimental Gemini 2.0 Flash became available to all Gemini users that day, alongside the launch of the Deep Research feature.⁠[18][18][18]

On July 21, 2025, Google DeepMind announced that an advanced version of Gemini Deep Think had perfectly solved five of the six problems at the 2025 International Mathematical Olympiad (IMO), scoring 35 points and achieving gold-medal level as graded by IMO coordinators using the same criteria as for students; the model answered directly in natural language and finished within the 4.5-hour competition time limit. The previous year, the combined AlphaProof and AlphaGeometry 2 systems had solved four problems and scored 28 points, reaching silver-medal level.⁠[19][19][19][19][19]

On November 18, 2025, Google launched Gemini 3; Gemini 3 Pro topped the LMArena leaderboard with an Elo score of 1501. In the launch post, Pichai said the Gemini app had more than 650 million monthly users and that 13 million developers had built with its generative models.⁠[20][20][20][20]

On September 30, 2026, Google DeepMind announced the frontier model Gemini 4 Argon, focused on long-horizon software engineering, knowledge work such as legal and finance, and cybersecurity defense. The model was first made available to a set of trusted cyber defenders through the Fairwind Program, and Google was also taking part in the US government’s voluntary process for pre-release model access, planning to open access to paid API customers, Google AI Ultra subscribers, and then more broadly to developers, enterprises, and consumers after strengthening safeguards. As of October 2026, Argon had not yet been made generally available.⁠[21][21][21][21]

Google DeepMind describes its mission as “to build AI responsibly to benefit humanity,” and believes that AI, and ultimately artificial general intelligence (AGI), has the potential to drive one of the greatest transformations in history. As of October 2026, the models and research listed on its official website included the Gemini family, the open model Gemma, the video model Veo, the music model Lyria, the world model Genie 3, Gemini Robotics, and scientific research projects such as AlphaFold, AlphaGenome, WeatherNext, and AlphaEvolve.⁠[1][1][1][1][1][1]

Google DeepMind: Official website: https://deepmind.google/ (opens in a new window).⁠[1]

X: Public account: https://x.com/googledeepmind (opens in a new window)⁠[22]

LinkedInLinkedInLinkedIn is a social networking platform for professionals. It was founded in 2002 by Reid Hoffman and others, launched in 2003, and listed on the New York Stock Exchange in 2011; after Microsoft acquired it for approximately $26.2 billion in 2016, it became a wholly owned Microsoft subsidiary. Its revenue comes from membership subscriptions, advertising, and recruitment solutions.Open the full entry: Public profile: https://www.linkedin.com/company/googledeepmind/ (opens in a new window)⁠[22]

YouTube: Public channel: https://www.youtube.com/@googledeepmind (opens in a new window)⁠[22]