← All companies

Frontier / Company profile

Google DeepMind

What Google DeepMind does

Google DeepMind is Google’s frontier AI lab and research-and-products organization, built on the combination of DeepMind and Google Brain and now positioned as the AI “engine room” for Google and Alphabet. On its own site, DeepMind describes itself as working on next-generation models and systems across Gemini (including multimodal agents), world models and physical AI (including robotics), and open models, with a strong emphasis on responsibility, safety, and scientific impact. Its strategy blends frontier research breakthroughs with deployment through Google’s products and developer platforms—most visibly through Gemini models and the Gemini app, plus model access through Google AI Studio and Gemini developer tooling.

DeepMind’s product direction has expanded from game-playing agents and foundation-model research into multimodal world models and embodied/robotics systems. On the lab side, it continues to pursue “AI for Science” themes such as protein structure prediction (AlphaFold) and newer research lines in climate and other scientific domains. On the deployment side, DeepMind’s frontier models are integrated across Google’s ecosystem: the organization points developers and users to Gemini (including specialized models such as Veo for video generation and Imagen for image generation) and to “Genie” world-model technology for interactive, simulated environments that can support agent training.

As of mid-August 2026, DeepMind is also undergoing a leadership transition described by Google/Alphabet executives: Demis Hassabis shifted from day-to-day CEO responsibilities to a chair and chief-scientist role, while Koray Kavukcuoglu stepped up to lead Google DeepMind as Senior Vice President, overseeing Gemini model development, frontier AI research, and Gemini app/developer teams.

News

Company record

Jul 30, 2026 · Official · Google DeepMind (deepmind.google)Gemini Robotics 2: whole-body intelligence for robots; announced as an intelligence layer for adaptable robots

DeepMind introduced Gemini Robotics 2, describing whole-body control, advanced dexterity, and multi-robot collaboration, and stated that the intelligence can run on-device while adapting to new robotic bodies in a few hours.

Jun 18, 2026 · Official · Google DeepMind (deepmind.google)Securing internal systems against increasingly capable AI agents: AI Control Roadmap

DeepMind described its AI Control Roadmap as a framework for building and managing advanced AI deployed within Google, treating untrusted AI agents as potential insider threats and focusing on securing internal systems.

May 19, 2026 · Official · Google DeepMind (deepmind.google)Co-Scientist: multi-agent AI partner to accelerate research; availability through Hypothesis Generation

DeepMind announced Co-Scientist, describing it as a coalition of specialized agents based on Gemini, and said it is made available to individual researchers through Hypothesis Generation.

Apr 22, 2026 · Official · Google DeepMind (deepmind.google)Partnering with industry leaders (Accenture, Bain & Company, BCG, Deloitte, McKinsey) to accelerate AI transformation

DeepMind described a partnership effort with major consulting firms to help bring frontier AI capabilities to organizations globally.

Apr 14, 2026 · Official · Google DeepMind (deepmind.google)Gemini Robotics-ER 1.6: enhanced embodied reasoning; developer availability via Gemini API and Google AI Studio

DeepMind released Gemini Robotics-ER 1.6, focusing on reasoning-first embodied capabilities such as spatial reasoning and multi-view understanding, and stated it is available to developers via the Gemini API and Google AI Studio.

Feb 18, 2026 · Official · Google DeepMind (deepmind.google)Accelerating discovery in India through AI-powered science and education (National Partnerships for AI)

DeepMind described its National Partnerships for AI initiative in India and stated it is working with government and other stakeholders to broaden access to frontier AI capabilities in sectors including science and education.

2026-02-?? · Official · Google DeepMind (deepmind.google)Gemini Deep Think: accelerating mathematical and scientific discovery

DeepMind described Gemini Deep Think as an AI system for scientific research and cited performance outcomes at competitive benchmarks and math/science discovery workflows.

Nov 19, 2025 · Official · Google DeepMind (deepmind.google)We’re expanding our presence in Singapore: opening a new Google DeepMind research lab for Asia-Pacific

DeepMind announced a new research lab in Singapore, describing plans to advance AI benefits and work with government and partners in the Asia-Pacific region.

Show 1 earlier updates
Source map · 6 recurring channels · 21 references

What it builds

Gemini

DeepMind describes Gemini as a family of intelligent agents and multimodal AI models, including Gemini Omni and specialized experiences such as audio, image, video, and reasoning features.

source ↗
Gemini Robotics

DeepMind introduced Gemini Robotics as AI models designed for robots, including a vision-language-action model for controlling robot actions and a spatial reasoning variant for embodied reasoning.

source ↗
Gemini Robotics-ER

DeepMind’s embodied-reasoning variant for robots, introduced alongside Gemini Robotics, focuses on spatial understanding and enabling roboticists to run programs using embodied reasoning.

source ↗
Genie (world models)

DeepMind describes its “world models & physical AI” line, including Genie 3 for generating and exploring interactive worlds.

source ↗
Gemma (open models)

DeepMind positions Gemma as its open models offering for building responsible AI applications at scale.

source ↗
Gemini app/platform access via Google AI Studio & Gemini API

DeepMind states that Gemini Robotics-ER updates are made available to developers via the Gemini API and Google AI Studio.

source ↗
Veo (video generation)

DeepMind lists Veo as a specialized model for generating cinematic video with audio.

source ↗
Imagen (image generation)

DeepMind lists Imagen as a specialized model for generating high-quality images from text.

source ↗
Lyria (music/audio)

DeepMind lists Lyria as a specialized model for generating high-fidelity music and audio.

source ↗
AlphaFold

DeepMind’s AlphaFold is used for predicting protein structures; DeepMind describes AlphaFold and highlights its protein-structure database availability in its science materials and launch posts.

source ↗
Co-Scientist (multi-agent research assistant)

DeepMind describes Co-Scientist as a multi-agent system to accelerate research, made available to individual researchers through a tool referred to as Hypothesis Generation.

source ↗

Milestones & partnerships