Google DeepMind

Google DeepMind

Overview

Google DeepMind represents a major consolidation of AI capabilities at Alphabet, combining the storied DeepMind research institute with Google's Brain team. DeepMind was founded in 2010 in London by [[Demis Hassabis]], Shane Legg, and Mustafa Suleyman, establishing itself as a leading AI research organization focused on advancing artificial general intelligence. Google acquired the company in 2014, and in 2023 formally merged DeepMind with Google Brain to create a unified frontier AI research operation under the Google DeepMind banner. This integration brings together decades of deep learning breakthroughs and gives the organization significant computing resources, enabling rapid iteration on increasingly capable models.

The organization's stated mission—"solve intelligence, then use that to solve everything else"—reflects an ambitious scope encompassing both pure research and applied breakthrough applications. Today, Google DeepMind operates as a core pillar of Alphabet's AI strategy, integrated directly into Google rather than as a separately valued subsidiary. With operations spanning London (original headquarters) and Mountain View, California, the lab pursues both frontier large language models and specialized research in domains like video generation, world modeling, and scientific discovery. 2026 has been an unusually productive year: Gemini 3.1 Pro shipped in February, Gemini 3.1 Ultra and Deep Think mode followed soon after, the Gemma 4 open-weight family launched in April, and Veo 4 — the new state-of-the-art video generation model — released alongside in April. AlphaEvolve has also reportedly been deployed inside Google's infrastructure for over a year, recovering ~0.7% of Google's worldwide compute continuously.

Key Details

  • Founded: 2010 (DeepMind, London); merged with Google Brain 2023
  • Founders: [[Demis Hassabis]], Shane Legg, Mustafa Suleyman
  • CEO: [[Demis Hassabis]]
  • Headquarters: London, UK (with major presence in Mountain View, CA)
  • Parent Company: Alphabet Inc.
  • Website: https://deepmind.google

Current Models

  • [[Gemini 3.1 Ultra]]: Top-of-line frontier model; 94.3% on GPQA Diamond (record); native multimodal reasoning; uses Chain-of-Verification at inference (Spring 2026)
  • [[Gemini 3.1 Pro]]: 1M+ context window; ARC-AGI-2 benchmark leader (77.1%); 2x+ reasoning improvement over Gemini 3 Pro; released February 19, 2026
  • Gemini 3.1 Deep Think — Optional deep-reasoning mode introduced February 2026
  • [[Gemini 3 Pro]]: Advanced general-purpose model with broad capabilities
  • [[Gemini 3 Flash]]: High-performance efficient model; GPQA Diamond 90.4%; released December 17, 2025; includes Agentic Vision capabilities
  • [[Gemini 2.5 Flash]]: Earlier efficient variant in the lineup
  • [[Gemini 2.5 Flash-Lite]]: Lightweight variant for resource-constrained deployments
  • [[Gemma 4]]: Open-weight model family; E2B, E4B, 31B, 26B A4B sizes; 256K context, native vision/audio, 140+ languages; Apache 2.0 license; released April 2, 2026
  • Veo 4: State-of-the-art video generation; up to 4K at 120 fps; 15–30s clip durations; storyboarding; advanced character consistency (April 2026)
  • Veo 3.1: Previous-generation video model
  • Genie 3: World model capable of planning and simulation
  • AlphaEvolve: Coding agent reportedly recovering ~0.7% of Google's worldwide compute continuously and discovering new mathematical structures
  • AlphaGenome: DNA research and genomics applications
  • Nano Banana 2: Image-generation/editing model introduced February 2026
  • Lyria 3: Music model introduced February 2026
  • TurboQuant: ICLR 2026-presented KV-cache compression algorithm using PolarQuant + Quantized Johnson–Lindenstrauss to reduce attention memory overhead

Key People

  • [[Demis Hassabis]]: CEO and co-founder; leading AI researcher and entrepreneur
  • [[Geoffrey Hinton]]: Distinguished researcher; worked at Google Brain; Turing Award winner (2018, shared); left Google in 2023 to speak openly about AI risks and safety concerns
  • Shane Legg: Co-founder of DeepMind
  • Mustafa Suleyman: Co-founder of DeepMind

Recent Developments

  • Pentagon Classified Deployment (May 1, 2026): Google cleared as one of seven (now eight) Big Tech AI vendors approved for IL-6/IL-7 classified Department of Defense networks via the GenAI.mil portal.
  • Gemma 4 Open-Weight Release (April 2, 2026): Apache 2.0 open-weight family with sizes E2B, E4B, 31B, and 26B A4B; 256K context window; native multimodal (vision + audio); 140+ language support; positioned for advanced reasoning and agentic workflows; optimized for NVIDIA Jetson Orin Nano through Blackwell, AMD ROCm, and Google Trillium/Ironwood TPUs.
  • Veo 4 Launch (April 2026): New flagship video generation model with 4K @ 120fps support, 15–30 second durations, multimodal input, and substantially improved character/temporal consistency. Available via Google AI Studio API.
  • AlphaEvolve Disclosure (May 2026): Reporting confirms AlphaEvolve has been continuously deployed inside Google's infrastructure for over a year, recovering ~0.7% of Google's worldwide compute resources while also discovering new mathematical structures — a striking validation of autonomous research agents at hyperscale.
  • TurboQuant at ICLR 2026 (May 2026): Two-step KV-cache compression algorithm targeting million-token context economics, with implications for both on-device inference and data-center cost structures.
  • Gemini 3.1 Pro Release (February 19, 2026): 1M-token context, 65K output tokens, 77.1% on ARC-AGI-2, 2x+ reasoning improvement over Gemini 3 Pro.
  • Gemini 3.1 Ultra: Achieved a record 94.3% on GPQA Diamond, surpassing prior internal builds; uses Chain-of-Verification at inference time, reportedly cutting hallucination rates in technical/scientific writing by 60%+.
  • ProducerAI Music Platform: Launched February 2026, extending creative applications.
  • Boston Dynamics Partnership: Strategic partnership on humanoid robotics integration.
  • DOE Genesis Project: U.S. Department of Energy collaboration on accelerated scientific discovery.

Why They Matter

Google DeepMind is one of the world's largest and most well-resourced AI research organizations, with direct access to Alphabet's vast computational infrastructure and billions in annual funding. The lab sets benchmark standards in frontier AI capabilities, particularly in reasoning, context windows, and video generation. Their models define much of the competitive landscape for large language models and multimodal systems. With Demis Hassabis leading the organization and deep talent including [[Geoffrey Hinton]]'s historical contributions, DeepMind bridges cutting-edge research with production deployment at Google scale. The 2026 cadence — Gemini 3.1 family, Gemma 4 open weights, Veo 4 video — combined with AlphaEvolve's quiet but enormous infrastructure impact (0.7% of Google's worldwide compute reclaimed) makes DeepMind the most multidimensional frontier lab today, simultaneously competitive in proprietary frontier scale, open-weight distribution, video generation, and autonomous research agents.

Last Updated

May 7, 2026

→ Back to Companies