Read this to see who stands behind Gemini and where the lab stands today. Google announced Gemini 4 Argon at the end of September 2026, yet only trusted cyber defenders in the Fairwind Program can use it, so developers wait. The lab's science record, from AlphaFold to WeatherNext 3, is the better-proven side of the business.
Google DeepMind is Alphabet's AI research lab, formed in April 2023 by merging Google Brain with DeepMind, which Demis Hassabis co-founded in London in 2010. It builds the Gemini model family, the open-weight Gemma models and AlphaFold, whose public database holds predicted structures for over 200 million proteins.
Founded: 2010 · HQ: London, UK · Team: 6,000-9,000 · CEO: Demis Hassabis · Funding: Acquired by Google in Jan 2014 for ~£400M (~$650M USD); wholly-owned Alphabet subsidiary (NASDAQ: GOOGL) since then
About Google DeepMind
Google DeepMind is the AI research lab of Alphabet and the group behind Google's Gemini models. DeepMind Technologies was founded in London in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman, with an interdisciplinary plan that mixed neuroscience, machine learning and engineering. Google bought the company in January 2014, and in April 2023 the lab merged with Google Brain, Google's other research team, to form Google DeepMind under CEO Demis Hassabis. Its own about page describes the merger as bringing two of the world's leading AI labs into a single team. On the Google side, Google is the parent that ships most of its output.
The research record explains why the lab is taken seriously. Its DQN system learned 49 Atari games from raw pixels, AlphaGo became the first program to beat a Go world champion in 2015, and AlphaZero and MuZero generalised the method to chess, shogi and other games. In 2020 AlphaFold2 won the CASP14 protein-structure challenge, and the AlphaFold Protein Structure Database, built with EMBL-EBI, now holds predictions for over 200 million proteins. DeepMind says more than 3 million researchers in over 190 countries use it. Demis Hassabis and John Jumper shared the 2024 Nobel Prize in Chemistry for that work, with David Baker. Nothing else in the lab's portfolio has a track record with outside validation like this.
The product side moves faster and is easier to get wrong. On 30 September 2026 Google announced Gemini 4 Argon, its new frontier model, and in the same post said it is rolling out only to trusted cyber defenders through the Fairwind Program. Developers, enterprises and consumers get it later, starting with paid API customers and Google AI Ultra subscribers, with no date given. Google states that the model raises the output limit to 1 million tokens, up from 64K, and reports 77.9% on DeepSWE v1.1, 68% on CWE-bench v1 (tied for first) and 91.7% on LVBench. Those scores are Google's own, run on its own setup, and nobody outside Google can reproduce them yet. What anyone can use today is the September 2026 generation: Gemini 3.8 Flash, the security-tuned Gemini 3.8 Flash Cyber and the voice model Gemini 3.8 Live.
Around the Gemini line sits a wide catalogue. Gemini Omni targets any-to-any generation, Veo 3.1 makes video with audio, and Lyria 3.5 makes music. Genie 3 builds interactive worlds, Gemini Robotics covers physical agents, and the open-weight Gemma 4 family includes Gemma 4 31B and Gemma 4 12B, which run from cloud servers down to laptops and phones. The science list keeps growing too: WeatherNext 3, AlphaGenome Atlas (predicted effects for all 9 billion single-letter variants in the human genome), AlphaEvolve for algorithm design and SIMA 2 for agents in 3D worlds.
Most readers meet this work through Google's own apps rather than the lab's site. Gemini is the consumer assistant, Gemini Notebook handles research notebooks, and Google AI Studio is the quickest route to the API. Enterprises usually reach the models through Google Vertex AI. For builders, Antigravity CLI, Gemini CLI and Jules turn the models into coding agents, and Gemini Spark is the delegating assistant inside the Gemini app. HokAI treats these as separate listings because the buying decision differs: the lab's page says who made the model, the product pages say what it costs and what it does for you.
Safety is a stated priority and a visible cost. DeepMind's Frontier Safety Framework sets capability thresholds that trigger extra safeguards before release, and the Argon announcement lists four areas of work: defending against misuse, resisting indirect prompt injection, monitoring the model's chain of thought for misalignment, and hardening the sandboxes used in testing. Google also says it is taking part in the U.S. government's voluntary process for pre-release model access. The phased release is the practical result, because the strongest model is withheld from almost everyone while those checks run. The lab also opened the DeepMind Institute in September 2026 as a platform for research on a world with AGI.
The other side of the story is availability and proof. A lab that holds back its flagship cannot win a buyer who needs a frontier model this week, and rivals ship generally available models that teams can adopt now, and OpenAI is the clearest example. For the other side in detail, read HokAI's profile of Anthropic first, then the one on xAI at this page. Vendor benchmark charts are also one tested configuration, so a team choosing between them should read an independent comparison such as Grok 4.7 vs Opus 5 vs GPT-6 Astra vs Gemini 3.8 and the explainer what Gemini is for. People who need open weights, on-device use or a scientific database get more from DeepMind than from most rivals, while people who need a product with a support team and a price list should look at the Google products listed above instead.
Hassabis is CEO, and Koray Kavukcuoglu, SVP of Google DeepMind and Google's Chief AI Architect, wrote the Argon announcement and leads the group building the generative models. DeepMind does not publish standalone accounts, so its size, budget and revenue are not separately known, and any figure quoted for them is an outside estimate.
Mission
Build AI responsibly to benefit humanity.
Products
- Gemini 4 Argon (Frontier model, announced 2026-09-30, limited to Fairwind Program partners): https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
- Gemini (Multimodal model family and consumer assistant): https://gemini.google.com
- Gemini API and Google AI Studio (Developer access to Gemini models): https://aistudio.google.com
- Gemma 4 (Open-weight model family for local and cloud use): https://deepmind.google/models/gemma/
- Gemini Omni (Any-to-any generation model): https://deepmind.google/models/gemini-omni/
- Veo (Video generation with audio): https://deepmind.google/models/veo/
- Lyria (Music and audio generation): https://deepmind.google/models/lyria/
- Genie 3 (Interactive world generation): https://deepmind.google/models/genie/
- Gemini Robotics (Models for physical robot control): https://deepmind.google/models/gemini-robotics/
- AlphaFold (Protein structure prediction and database): https://deepmind.google/science/alphafold/
- WeatherNext (AI weather forecasting): https://deepmind.google/science/weathernext/
Google DeepMind Models on HokAI
Compliance
SOC 1 Type II, SOC 2 Type II, SOC 3, ISO 27001, ISO 27017, ISO 27018, ISO 27701, ISO 42001, HIPAA-eligible, FedRAMP High, PCI DSS, GDPR
Links
Frequently Asked Questions
Who owns Google DeepMind, and has it raised outside funding?
Google DeepMind does not raise outside money. DeepMind Technologies took early private funding, reportedly including Horizons Ventures and Founders Fund, before Google acquired it in January 2014 for a reported 400 million pounds (about 650 million US dollars at the time). Since then it has been wholly owned by Google and Alphabet (NASDAQ: GOOGL), and it publishes no standalone valuation or budget.
What products does Google DeepMind make in 2026?
The Gemini model family is the core: Gemini 4 Argon was announced on 30 September 2026 but is limited to Fairwind Program partners, while Gemini 3.8 Flash and Gemini 3.8 Live are available now. Google lists introductory Argon API pricing of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Gemma 4 open weights are free to download, and the AlphaFold database is free for researchers. Consumer plan prices belong to the Gemini product page, not the lab.
How does Google DeepMind handle data privacy and compliance?
Enterprise use of the models runs through Google Cloud, whose compliance page lists SOC 1, SOC 2 and SOC 3 reports, ISO 27001, 27017, 27018, 27701 and 42001, HIPAA support, FedRAMP, PCI DSS and GDPR. Those certifications belong to the cloud platform rather than to the research lab itself. Retention terms differ by product, so read the Vertex AI or Gemini API terms for the exact plan before sending sensitive data.
How does Google DeepMind compare with OpenAI and Anthropic?
OpenAI and Anthropic are the closest frontier rivals, and xAI and Meta AI compete on models and open weights. DeepMind wins on research breadth (protein structure, weather, genomics), open Gemma weights and distribution through Google's own apps. It trails when a rival has a frontier model on general sale and DeepMind's own flagship is still restricted to partners.
Which Google DeepMind models can you try today?
Most people start in the Gemini app or in Google AI Studio, which gives free API access with usage limits. Companies usually go through Vertex AI on Google Cloud for service terms and regional controls. Gemma 4 can be downloaded and run locally, and Gemini 4 Argon is not open to the public yet, so plan around Gemini 3.8 until Google announces wider access.
Top Alternatives
HokAI guides covering Google DeepMind
- Best AI for Generating Pictures of People in 2026: Best AI for generating pictures of people in 2026: prices from $0.0336 an image, keeping one face across scenes, rights, watermarks and who should skip it.
- Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5: Cost Per Real Task: Argon, GPT-6 Astra and Claude Opus 5.5 compared on what a finished task costs, not the token price, plus the 1 December price rise on Argon. Updated Oct 2026.
- Grok 4.7 vs Claude Opus 5 vs GPT-6 Astra vs Gemini 3.8 Flash: What the Launch Numbers Don't Tell You: Grok 4.7, Claude Opus 5, GPT-6 Astra and Gemini 3.8 Flash all launched within seven weeks. Here is what their price, speed and benchmark numbers actually mean.
- What Gemini Omni Flash Is For: Google's Default Video Model, What It Costs, and Where Veo 3.1 Still Wins: Google now calls Gemini Omni Flash its default video model. See what it costs next to Veo 3.1, what the Gemini app plans include, and where Veo still wins.
- Project Glasswing Explained: What Claude Mythos Found Inside Epic, and Who Else Gets Access: What Project Glasswing is, who is in it, what Epic found with Claude Mythos, and how Google Fairwind and OpenAI Codex Security compare on access and price.
- AMD Buys World Labs for $8.2B: What It Means for Marble Users: AMD will buy World Labs for about $8.2B in stock, yet neither side mentions Marble. What subscribers and API users should export, cap and test before closing.