What is Google DeepMind? It is Google's main artificial intelligence lab, a unit of Alphabet based in London, and it builds the Gemini models behind the Gemini app, AI Mode in Search and the Gemini API. It also runs science projects such as AlphaFold, the protein structure system that won a share of the 2024 Nobel Prize in Chemistry, and world models such as Genie. You don't sign up for "DeepMind" itself: you use what it builds, and several of those products have free versions.
This guide explains where the lab came from, who runs it after the August 2026 leadership change, what it makes today, which parts you can use for free, and how it relates to Gemini. Facts were checked on November 2, 2026 against Google DeepMind's own pages, Google's blog, the Nobel Prize announcement and reporting from Reuters, The Guardian and Axios.
Google DeepMind at a glance
| Question | Short answer |
|---|---|
| What is it? | Google's AI research and model lab, owned by Alphabet |
| Founded | 2010 in London, as DeepMind |
| Founders | Demis Hassabis, Shane Legg and Mustafa Suleyman |
| Part of Google since | January 2014 (acquisition) |
| Current form | Created in April 2023 by merging DeepMind with the Google Brain team |
| Who runs it | Koray Kavukcuoglu (SVP) since August 2026; Demis Hassabis is Chair |
| Best known for | Gemini, AlphaGo, AlphaFold, WaveNet, Genie |
| Can you use it? | Yes, through products such as the Gemini app, Google AI Studio and AlphaFold Server |
| Our directory page | Google DeepMind on intelsi.si |
What Google DeepMind does
Google DeepMind's About page states its mission as building AI "responsibly to benefit humanity" and says AI, "and ultimately artificial general intelligence (AGI)," could drive one of the greatest transformations in history. In practice the lab does three kinds of work.
It trains Google's frontier models. The Google DeepMind models page lists the current lineup: Gemini 4 Carbon (the newest frontier model, October 2026), Gemini 3.8 Flash, Gemini Audio, Gemini Omni for video, Nano Banana for images, Veo for video, Lyria for music, Gemini Embedding and the Gemma open models. Hassabis wrote in August 2026 that Gemini powers features across Google, including AI Mode and AI Overviews in Search.
It does science research. AlphaFold predicts the 3D shape of proteins, and the lab has also worked on weather forecasting, faster sorting algorithms (AlphaDev) and controlling plasma in fusion reactors, according to its About page.
It builds world models and robotics systems. Genie 4, announced on October 27, 2026, generates interactive worlds you can move through, and Gemini Robotics 2 is pitched as the intelligence layer for robots.
The lab also publishes safety research and runs Google's Frontier Safety Framework, its rulebook for testing models for dangerous abilities before release. Our Preparedness Framework vs Responsible Scaling Policy guide compares it with the OpenAI and Anthropic versions.
Is DeepMind part of Google?
Yes. DeepMind started as an independent London startup in 2010. Google agreed to buy it in January 2014; Google did not disclose the price, which Reuters reported as $400 million citing Re/code and The Guardian reported as £400 million (about $650 million).
For nine years DeepMind and Google's own research group, Google Brain, worked as separate teams. On April 20, 2023, Sundar Pichai announced that the two would become one unit called Google DeepMind, led by Hassabis. That merger matters because Google Brain invented the Transformer architecture in 2017, the design behind almost every large language model, while DeepMind brought deep reinforcement learning and AlphaFold. Today the lab sits inside Alphabet and reports to Pichai.
A short history of DeepMind
| Year | Milestone (per Google DeepMind and the sources cited) |
|---|---|
| 2010 | DeepMind founded in London |
| 2014 | Google acquires DeepMind |
| 2015 | AlphaGo unveiled; it became the first program to defeat a Go world champion |
| 2017 | Google Brain invents the Transformer |
| 2019 | AlphaStar beats a top professional StarCraft II player |
| 2020 | AlphaFold presented; CASP14 organizers call it a solution to protein folding |
| 2023 | DeepMind and Google Brain merge into Google DeepMind |
| 2024 | Hassabis and John Jumper share the Nobel Prize in Chemistry |
| 2026 | Gemini 4 Argon and Carbon released; Hassabis becomes Chair |
Two early results set the lab's reputation. DQN learned to play 49 Atari games from raw screen pixels, and AlphaGo beat a Go world champion, a goal many researchers thought was years away. The lab's WaveNet speech model later became the voice of Google Assistant.
The biggest scientific result came with AlphaFold. The Nobel Prize press release of October 9, 2024 gave half the chemistry prize jointly to Hassabis and Jumper "for protein structure prediction," noting that AlphaFold2 has been used by more than two million people in 190 countries.
Who runs Google DeepMind now
On August 5, 2026, Google announced a leadership change. Demis Hassabis handed over day to day running of the lab and became Chair of Google DeepMind and Chief Scientist of Alphabet, while continuing to lead Isomorphic Labs, the Alphabet drug discovery company built on AlphaFold. Koray Kavukcuoglu, previously the lab's chief technology officer and Google's Chief AI Architect, became SVP of Google DeepMind, reporting to Pichai and overseeing Gemini model development, frontier research and the Gemini app and developer teams.
In his note to staff, Hassabis wrote that he has worked toward AGI his whole life and now feels "it is close at hand." The same day, Axios reported that Jeff Dean was leaving Google to start a public benefit company with Sanjay Ghemawat, joined by DeepMind vice president Oriol Vinyals and Google Brain cofounder Quoc Le, and that Alphabet shares fell more than 4% on the news.
Cofounder Mustafa Suleyman left long ago: he went on to cofound Inflection and in March 2024 joined Microsoft to lead Copilot, so one DeepMind founder now runs a rival assistant.
Is Google DeepMind the same as Gemini?
No, though the names are tied together. Google DeepMind is the lab. Gemini is two things it builds:
- Gemini models, the family of large models (currently Gemini 4 Argon, Gemini 4 Carbon and Gemini 3.8 Flash among others).
- The Gemini app, Google's assistant, which Pichai said in August 2026 had passed 950 million monthly users. Since the August change, Kavukcuoglu oversees the Gemini app team as well as the models.
A simple way to think about it: DeepMind is to Gemini roughly what OpenAI is to ChatGPT, except DeepMind is a unit of a larger company rather than a standalone firm. If you want to compare the assistants themselves, read our ChatGPT Plus vs Claude Pro vs Google AI Pro guide. For the two newest models, see Gemini 4 Argon release date and Gemini 4 Carbon explained.
Is Google DeepMind free?
Many DeepMind products have a free option, but the newest and heaviest ones sit behind paid plans or vetted access.
| Product | Free? | What Google says (checked Nov 2, 2026) |
|---|---|---|
| Gemini app | Free tier | Free plan includes the Flash-Lite model, image generation, Deep Research and Gemini Live |
| Gemini 4 Argon in the app | Paid | Included in Google AI Pro and Ultra |
| Google AI Studio and Gemini API | Free tier | Free tier with no token cost; paid plan for higher limits and the most advanced models |
| AlphaFold Protein Structure Database | Free | Over 200 million predictions, free and open |
| AlphaFold Server | Free, noncommercial only | For individuals and noncommercial organizations, with a Google account |
| Project Genie (Genie 3) | Paid | Google AI Ultra only, 18 and over |
| Genie 4 | Not yet | Research preview for trusted testers; waitlist open |
| Gemini 4 Carbon | Not for the public | Vetted cyber defenders first; no public date |
The free Gemini plan details and the Ultra prices ($99.99 a month for the 5x tier, $199.99 for the 20x tier) come from Google's Gemini subscriptions page. The AlphaFold database figure comes from EMBL-EBI's AlphaFold DB page, run with Google DeepMind, and the server rules from AlphaFold Server, which bars commercial use and using outputs to train similar structure prediction models.
For developers, Google AI Studio is the easiest way to try Gemini models in a browser and get an API key. Our Google AI Studio vs Gemini comparison explains when to use the workbench and when the app is enough.
How to use Google DeepMind's Genie
Genie is the product people most often ask how to "use," because it lives outside the Gemini app. Today there are two versions:
- Project Genie (Genie 3). Google's Labs help page says you sign in at labs.google with a personal Google account that has Google AI Ultra, describe a world and a character (or start from a Street View location), pick first or third person, and explore with the keyboard. Each world lasts 60 seconds, and it does not use your AI credits while it is an early research prototype.
- Genie 4. The Genie 4 announcement adds native audio, spoken dialogue and text prompts that change a scene while it runs. It is in a research preview for trusted testers and is "coming soon" to Ultra subscribers in Project Genie and to early access partners in the Gemini API and AI Studio. Google lists drift over time, text rendering and sessions of only a few minutes as limits.
Google calls simulating the world "a crucial step toward AGI," because it gives robots and agents a safe place to train.
Pros and cons of Google DeepMind's approach
Strengths
- A long record of science results, including a Nobel Prize, not only chatbots.
- Google's scale: its models reach users through Search, Android, Gmail and the Gemini app.
- Free access to major research tools such as the AlphaFold database.
- A published safety framework and model cards for flagship releases such as Gemini 4 Argon.
Weaknesses and open questions
- Newest models often launch to small vetted groups first, so ordinary users wait.
- Product names are confusing: "Gemini" covers models, an app and an API.
- The August 2026 reshuffle followed staff departures, and Axios reported that Gemini 3.5 Pro was running months behind.
- Strong world model and Genie features are tied to the most expensive Ultra plan.
What Google DeepMind says about AGI and safety
DeepMind is one of the few labs that has always named AGI as its goal. Its researchers wrote the "Levels of AGI" paper that grades systems from Emerging to Superhuman, which we cover in superintelligence vs AGI and are AI agents AGI. On safety, its Frontier Safety Framework sets Critical Capability Levels that trigger extra protections, and its Gemini 4 Argon model card reported that some loss of control warning levels could not be ruled out. Our guide to how autonomous frontier AI models are covers what that card found.
FAQ
Is Google DeepMind free to use?
Partly. The Gemini app has a free plan, Google AI Studio has a free API tier, and the AlphaFold database is free for everyone. AlphaFold Server is free for noncommercial use. Project Genie needs Google AI Ultra, and Gemini 4 Carbon is limited to vetted partners.
Is DeepMind part of Google?
Yes. Google bought DeepMind in 2014 and merged it with Google Brain in April 2023 to form Google DeepMind, a unit of Alphabet that reports to CEO Sundar Pichai.
Is Google DeepMind the same as Gemini?
No. Google DeepMind is the lab; Gemini is the model family and the assistant app it builds. Using the Gemini app is the most common way people use DeepMind's work.
Who is the CEO of Google DeepMind?
Google's August 5, 2026 announcement did not name a new CEO. Koray Kavukcuoglu leads the lab as SVP of Google DeepMind, and Demis Hassabis is Chair of Google DeepMind and Chief Scientist of Alphabet.
What does Google DeepMind do besides Gemini?
Science and research projects such as AlphaFold for protein structures, the Genie world models, Gemini Robotics, music and video models (Lyria and Veo), and the Gemma open models. Its drug discovery sibling, Isomorphic Labs, is led by Hassabis.
Where is Google DeepMind based?
Its headquarters is in London. Hassabis's August 2026 note mentions the lab's new Platform 37 offices in London.
