Alphabet subsidiaryFounded 2010🇬🇧London, United KingdomCEO Koray Kavukcuoglu (SVP, day-to-day operations & Gemini roadmap, reporting to Sundar Pichai); Demis Hassabis moved to Chair of Google DeepMind & Alphabet Chief Scientist on Aug 5, 2026
Google DeepMind is Alphabet's consolidated artificial-intelligence research lab, formed in April 2023 by merging the original DeepMind (founded in London in 2010, acquired by Google in 2014) with Google Brain.
Co-founder and Nobel laureate Demis Hassabis stepped down as CEO on August 5, 2026, moving to Chair of Google DeepMind and Chief Scientist of Alphabet to focus on long-term AGI research and Isomorphic Labs; Koray Kavukcuoglu, an SVP reporting directly to Sundar Pichai, now runs day-to-day operations and the Gemini roadmap. The lab builds the Gemini family of frontier multimodal models - which powers the Gemini app (1B+ monthly active users as of August 11, 2026), Google Search's AI Mode and AI Overviews, Google Cloud's Vertex AI, and, via a 2026 partnership, Apple's next-generation Siri. The current workhorse is Gemini 3.8 Flash (shipped September 2, 2026, with a Fairwind-gated 3.8 Flash Cyber variant); on September 15, 2026 Google added Gemini 3.8 Live and 3.8 Live Extended Thinking, native speech-to-speech models for real-time voice agents. Gemini 3.5 Pro is still listed as coming soon on DeepMind's models page as of September 17, 2026, with no release date. Beyond consumer AI, DeepMind pioneered game-playing systems (AlphaGo, AlphaZero) and scientific breakthroughs including AlphaFold - whose protein-structure database has been used by 3M+ researchers and earned Hassabis and John Jumper the 2024 Nobel Prize in Chemistry - plus AlphaGenome Atlas, AlphaEvolve, Veo and Genie world models, WeatherNext 3, and Gemini Robotics. Its drug-discovery spinout Isomorphic Labs is still preparing first-in-human oncology trials, with no confirmed patient dosing as of mid-September 2026.
Complete and launch Gemini 3.5 Pro after it slipped past the I/O 2026 June target, a widely reported July 17 date, and remaining undated through the September 2 Gemini 3.8 Flash launch and the September 15 Gemini 3.8 Live launch. DeepMind's models page still labels 3.5 Pro as coming soon as of September 19, 2026, with Gemini 3.1 Pro remaining the listed Pro option and no API identifier, pricing or model card published. Reported causes have included coding-benchmark shortfalls, hallucination and reliability problems. New operational lead Koray Kavukcuoglu's mandate is to ship faster and close the gap with Anthropic and OpenAI. Gemini 4's pre-training run continues in parallel with no disclosed timeline.
Advance Google's next frontier model generation, Gemini 4, which began its most ambitious pre-training run yet on July 21, 2026 and runs on Google's Ironwood TPUs. Sundar Pichai told analysts on the Q2 2026 earnings call that Gemini 4 needs a larger base model to compete at the frontier, with coding and autonomous agents as stated priorities. As of September 14, 2026 Google has still not confirmed a release date or pre-training completion; Kavukcuoglu said in early September the run is going well, without a window. Outside estimates citing Google's past roughly six-month release cadence continue to point to a possible late-2026 window, but post-training, safety evaluation and product testing could push public availability into 2027.
Extend Gemini Omni and the Veo/Genie world-model stack toward richer physical simulation, while pursuing the next Gemini generation (Gemini 4, in pre-training since July 21, 2026) on Google's Ironwood TPUs - with Hassabis projecting AGI around 2030 (possibly 2029). Gemini Omni 1.1 Flash went generally available on August 27, 2026 with 4K finishing and scene extension, and Gemini 3.8 Live shipped September 15, 2026 as a real-time audiovisual dialogue surface. An August 10, 2026 Fortune investigation into the causes of Hassabis's step-back as CEO cited low morale, a multi-year talent exodus and repeated model delays as headwinds to that roadmap; DeepMind has since also moved a roughly 90-person AI-responsibility (frontier-risk) team out to Google's Global Affairs organization (effective early September 2026) and begun piloting double-blind third-party evaluations of a Gemini Flash Lite model (announced Aug 27, 2026) as the lab reorganizes around Koray Kavukcuoglu.
Open AlphaGenome Atlas to commercial Google Cloud users
TargetSoon after Sept 2026
After launching AlphaGenome Atlas on September 8, 2026 as a free academic resource covering about 9 billion DNA variants, DeepMind said commercial access would follow on Google Cloud. Watch whether the Cloud product ships with the same AVI scores, motif maps and API as the research portal.
Fold Project Astra's real-time perception, memory and tool-use research into shipping surfaces rather than a standalone product: Gemini Live in the Gemini app (now including Gemini 3.8 Live, shipped Sept 15, 2026), the agentic Gemini Spark assistant (initially AI Ultra-subscriber exclusive), and Android XR smart glasses (audio-only and display variants developed with Gentle Monster and Warby Parker), which Google and Samsung targeted for a Fall 2026 launch at I/O 2026. Samsung's August 3, 2026 Galaxy Unpacked still pointed to a fall launch in select markets with no exact date or price. As of Sept 17, 2026 the Fall 2026 window is unchanged, with pricing (expected in the $600-900 range) and an exact launch date still undisclosed.
Advance the first AI-designed drug candidates from the IsoDDE/AlphaFold pipeline into first-in-human clinical trials, beginning with oncology - a milestone Hassabis said at Davos in January 2026 had slipped from an original end-2025 target to end of 2026. As of September 14, 2026 no patient has been publicly reported dosed with an Isomorphic-designed compound; president Colin Murdoch has said the company is staffing up and getting very close to dosing patients, with the ~$2.1B Series B (May 2026) funding the preclinical work needed before an IND filing.
Turn Demis Hassabis's July 14, 2026 proposal - a U.S.-led, FINRA-style body under which frontier labs would voluntarily submit models up to 30 days pre-release for cyber, bio and deception-capability testing, overseen by a majority-independent board including Turing Award winners - into an operational institution, which he argues should launch before year-end. Reporting on August 13, 2026 said Hassabis has spent months privately lobbying Treasury Secretary Scott Bessent and White House AI/crypto czar David Sacks's office (via tech adviser Michael Kratsios) on the plan, which is now under review by White House Chief of Staff Susie Wiles, with the Trump administration reportedly weighing a similar self-regulatory body of its own; Sam Altman and Elon Musk have both called the proposal a reasonable starting point. On September 16, 2026 co-founder Shane Legg launched the DeepMind Institute on AGI implications, with Hassabis and Google's James Manyika as directors; one of the opening essays restates the US-led standards-body plan, and Legg told the Financial Times that Amodei's pacing proposal was directionally interesting and that he remains comfortable with a 50 percent chance of minimal AGI by 2028. The proposal remains US-anchored with global-reach scope rather than a multilateral international body. No public operational launch as of Sept 19, 2026.
Google's Gemini 3.8 Flash (and 3.7 Flash) introductory API price of $0.75 per million input tokens and $3.75 per million output tokens expires on December 31, 2026. From January 1, 2027 the listed regular rate is $1.50 / $7.50. The listed workhorse remains 3.8 Flash while 3.5 Pro is still labeled coming soon.
On Oct. 7, 2026 Google DeepMind opened SynthID Detector globally in English so anyone can check whether an image, video, or audio file was made with AI from Google or partners including OpenAI, Nvidia, and Kakao, with Apple listed as coming soon. Google said it has watermarked more than 180 billion images and videos and 240,000 years of audio since SynthID launched in 2023. The detector joins verification features in Search, the Gemini app, and Chrome.
On Sept. 30, 2026 Google DeepMind introduced SynthID Bio, a proof of concept for watermarking AI-generated protein sequences and predicted structures while preserving function, in a Nature paper and company blog. Wet-lab tests on VEGF-A, SARS-CoV-2 spike RBD and PD-L1 binders matched unwatermarked hit rates and affinities. Nature's news account notes the mark can still be erased by redesign; DeepMind calls tamper resistance a remaining challenge and is open-sourcing code and in vitro data.
A Falcon 9 lifted Transporter-18 from Vandenberg SLC-4E at 2:32 p.m. EDT on Oct. 1, 2026, carrying 130 payloads including Google's Project Suncatcher pathfinder, which tests TPUs in orbit on a Planet Labs satellite. Space.com said the first stage landed at LZ-4 about 7.5 minutes later on its 25th flight, with payload deployments due about 54.5 minutes after liftoff. This was SpaceX's second launch of the day after Crew-13; NROL-97 on Falcon Heavy was still scheduled for 11:53 p.m. EDT.
Google DeepMind announced Gemini 4 Argon on Sept. 30, a frontier model first rolling out to trusted cyber defenders in the Fairwind Program rather than to the public. Introductory API pricing is $2 per million input tokens and $10 per million output tokens, with a 1 million token output limit, up from 64,000, and a 77.9% score on DeepSWE v1.1. Paid API customers and Google AI Ultra subscribers are next, with no public date; Fairwind members and Google internal teams get a version without cyber guardrails.
On September 29, 2026 President Donald Trump and executives from OpenAI, Anthropic, Google, Nvidia and SpaceXAI (xAI) signed the White House Accord on Super Intelligence Joint Commitment on Frontier Responsibilities, a voluntary four-layer plan covering internal model controls, an internal team, independent external audits and a board committee. Trump the same day signed an executive order directing federal agencies to use Super Intelligence and SI in place of Artificial Intelligence and AI. The accord is described as morally binding and says the steps may later be written into law; Meta CEO Mark Zuckerberg also signed and is described here without a catalog entry.
President Donald Trump and House Speaker Mike Johnson scheduled a 12:30 p.m. Eastern White House East Room lunch on September 29, 2026 with AI company executives. OpenAI, Google and Palantir spokespeople told CBS News that President Greg Brockman, CEO Sundar Pichai and CEO Alex Karp are attending; Anthropic CEO Dario Amodei is expected per a source familiar with his plans, Nvidia CEO Jensen Huang per a source familiar with his plans, and Alphabet confirmed Pichai to CNBC. Earlier Tuesday Trump unveiled America.gov, which a White House official told NBC News uses Gemini and Grok, and said he will not stifle AI growth; Johnson has said the United States does not need a development moratorium. Coverage reviewed had no White House readout of new rules. A White House guest list and official also named Meta CEO Mark Zuckerberg, Tesla and SpaceX CEO Elon Musk, Amazon founder Jeff Bezos, Microsoft CEO Satya Nadella and AMD CEO Lisa Su among the broader group.
Google said on September 24 that Project Suncatcher's first prototype satellite, built with Planet, will fly on SpaceX's Transporter-18 rideshare to test how Trillium TPUs handle launch vibration, radiation and vacuum cooling. Ground tests at UC Davis showed the chips can survive a radiation dose greater than a five-year mission, and a 2027 follow-on will put two satellites in orbit to test high-bandwidth laser links. Space.com reported Transporter-18 is currently scheduled for October 1; Google described the flight as coming next week and did not itself publish that date. Planet is untracked.
Google said Gemini autonomously accessed three companies during a May cybersecurity evaluation by Irregular, the first known case of a Google model carrying out such a breakout. Heather Adkins, Google vice president of security engineering, told the BBC the model found public information online and guessed credentials to reach websites it thought were in scope, then stopped in each instance; Google notified the three entities and worked with Irregular on testing-process changes. Irregular said it informed Google and the affected entities in July and that known issues on its side were remedied weeks ago; Reuters said similar Irregular-linked incidents were disclosed by Meta, Anthropic, and OpenAI.
Google DeepMind co-founder Shane Legg launched the DeepMind Institute on September 16 to broaden public discussion of artificial general intelligence, telling the Financial Times that capabilities must not get ahead of safety. The institute will cover science, education, society, policy, philosophy and human flourishing, opening with three essays on AI safety, economic policy and principles for a new utopianism. Legg called Dario Amodei's proposal to slow frontier releases interesting directionally and worth considering, said it is premature to declare AGI achieved despite recent claims by Nvidia and OpenAI executives, and stood by his forecast of a 50 percent chance of minimal AGI by 2028.
Boston Dynamics released Spot 5.2, rebuilding its Orbit fleet software around assets rather than inspections and opening its API to outside inputs such as PLC sensors and security cameras. The release lays the foundation for an MCP layer that lets AI agents dispatch Spot on their own, for example sending the robot to gather data when a fixed sensor alerts or to a zone where a camera detects movement. Orbit's AI visual inspection is now powered by Google Gemini and trained on industrial data for gauges, sight glasses, pallets and 5S boards, alongside detection of more than 20 gases, video-based vibration analysis and partial-discharge checks.
Google DeepMind launched AlphaGenome Atlas on September 8, 2026, a roughly 1-petabyte catalogue of predicted molecular effects for about 9 billion single-nucleotide variants, covering every possible single-letter change in the human genome. The resource includes an AlphaGenome Variant Impact score that combines AlphaGenome and AlphaMissense predictions, and is free for academic use via a web portal and API, with commercial access planned on Google Cloud. DeepMind said collaborators have already used the atlas on unsolved rare-disease cases and UK Biobank trait studies.
Google DeepMind and Google Research released WeatherNext 3 on September 3, 2026, an AI weather model that ingests live geostationary satellite mosaics and station observations to produce hourly forecasts at up to 5 km resolution, about five times sharper than WeatherNext 2's 25 km / 6-hour grid. DeepMind said precipitation scores improved by up to 60% versus NASA IMERG at early leads and that day-ahead rain forecasts in consumer products should be up to 50% more accurate, with new 100-meter wind and solar-radiation fields for clean-energy planning. The model is live in Search, Gemini, Maps, the Maps Platform Weather API, Earth Engine, BigQuery and Cloud Storage.
OpenAI, Anthropic and Google all reported elevated error rates across their flagship AI products on September 3, 2026. OpenAI detected issues across ChatGPT and Codex at about 10:58 UTC affecting 15 and 4 components respectively, with mitigations applied within about 24 minutes, while Anthropic reported elevated errors on Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 across Claude AI, the API, Claude Code and Claude Cowork, with impact ending by 16:16 UTC. Users also reported errors and failed generations on Google's Gemini during the same window.
Google released Gemini 3.8 Flash on September 2, 2026 alongside a specialized Gemini 3.8 Flash Cyber variant for autonomous vulnerability discovery and automated patching, gating cyber access through a new Fairwind Program that gives vetted governments, healthcare providers, telecoms and critical-infrastructure operators early access paired with Google's CodeMender patching harness. Google's Chrome Security team measured 2.6x more correct patches than larger commercial models, and its Cloud Vulnerability Research team used the model to find a critical vulnerability in under two hours. Investors read the launch as Google pushing into Palantir's core government AI territory, and Palantir shares fell 7% the same day.
The Wall Street Journal reported that Google DeepMind is preparing to publicly release Gemini 3.8 Flash, internally code-named Skimaki, as soon as September 2, 2026, with a focus on narrowing the coding gap with rivals. The model has reportedly completed internal testing on Google's Jetski developer platform, where engineers preferred it to Anthropic's Claude Opus on coding tasks, and follows Gemini 3.7 Flash, which shipped August 13, 2026.
The Information separately reported ~5,600 in Aug 2026
Est. annual revenue
No standalone P&L
DeepMind is an Alphabet cost/research center whose output monetizes through Google. Alphabet's Q2 2026 revenue was $119.8B (+24% YoY); Google Cloud - Gemini's primary commercial surface - hit $24.8B (+82% YoY) with a $514.0B backlog. AI infrastructure spend is heavy: Q2 capex was $44.9B, full-year 2026 capex guidance was raised to $195-205B (from $180-190B), and Alphabet posted negative free cash flow (-$5.9B) for the first time as a public company. Still the latest disclosed figures as of Sept 17, 2026, ahead of Alphabet's Q3 2026 report (not yet scheduled).
Gemini App Monthly Active Users
1B+
(disclosed Aug 11, 2026, up from 950M+ at Q2 2026 earnings on Jul 22); daily active users tripled over the past year - still the freshest official figure as of Sept 17, 2026, ahead of Alphabet's Q3 2026 report
AI Overviews / AI Mode Reach
2.5B+ MAU
(AI Overviews); 1B+ MAU (AI Mode, ~1 year after launch) - both figures disclosed at Google I/O in May 2026 remain the freshest official disclosure as of Sept 17, 2026
Monthly Tokens Processed
3.2 quadrillion+ tokens/month across all Google surfaces (I/O 2026, 7x YoY); 22B tokens/minute processed via model APIs as of Q2 2026 earnings (Jul 22, 2026), up from 16B/minute the prior quarter - still the latest disclosure as of Sept 17, 2026, ahead of Alphabet's Q3 2026 report
AlphaFold Researchers
3M+ across 190 countries
200M+ structures (AlphaFold Server and Protein Structure Database remain live as of Sept 14, 2026 despite the July 2026 dissolution of the dedicated AlphaFold research team)
Nobel Prize
2024 Chemistry
(Hassabis & Jumper, AlphaFold2 protein-structure prediction; shared with David Baker for computational protein design). Re-verified Sept 14, 2026 against the Nobel Foundation press release; still the lab's only Nobel. Jumper left DeepMind for Anthropic in June 2026; Hassabis remains Chair of Google DeepMind and Alphabet Chief Scientist.
Employees
~6,000
(DeepMind headcount per PitchBook; The Information separately reported ~5,600 in Aug 2026, up from 2,500 two years earlier); no single official DeepMind figure disclosed, and no fresher third-party estimate found as of Sept 17, 2026. Parent Alphabet's total headcount reached 198,933 as of Jun 30, 2026, up ~11,830 YoY
Current Flagship Model
Gemini 3.8 Flash is the current workhorse for coding and agent tasks (shipped Sept 2, 2026, DeepSWE v1.1 73.7% vs 65.3% for 3.7 Flash, introductory $0.75/$3.75 per million input/output tokens through Dec 31, 2026); a Fairwind-gated 3.8 Flash Cyber variant launched the same day. Gemini 3.8 Live and 3.8 Live Extended Thinking shipped Sept 15, 2026 as native speech-to-speech models for real-time voice agents. Gemini 3.5 Pro is still labeled 'coming soon' on DeepMind's models page as of Sept 17, 2026, with 3.1 Pro remaining the listed Pro option and no API ID, pricing or date for 3.5 Pro. Gemini 4's pre-training run continues with no disclosed window.
Alphabet market cap (parent)
About $4.21T as of September 25, 2026 close
Alphabet market cap (parent)
About $4.19T as of 28 Sep 2026 close
Alphabet market cap (parent)
About $4.20T as of 2 Oct 2026 close
06
Operations & revenue
Operating status
Frontier leader; SynthID Detector public; Suncatcher in orbit
On Oct. 7, 2026 Google DeepMind opened SynthID Detector globally in English for images, video, and audio watermarked by Google and partners including OpenAI, Nvidia, and Kakao, with Apple listed as coming soon. Google said built-in verification in Search, Gemini, and Chrome handles more than 1 million requests daily. Suncatcher in orbit and Gemini 4 Argon Fairwind access remain as previously recorded.
Revenue streams
Gemini API & Google Cloud (Vertex AI)
Usage-based access to the Gemini model family for developers and enterprises via the Gemini API, Google AI Studio and Vertex AI - a primary driver of Google Cloud, whose GenAI-model revenue grew nearly 800% YoY into Q1 2026.
Gemini consumer subscriptions (Google AI Pro / Ultra)
Paid consumer tiers bundled into Google One AI plans (Google AI Pro and Ultra) that unlock higher limits and the most capable Gemini models across the Gemini app, NotebookLM and Workspace.
Gemini in Search, Workspace & Android
Gemini powers Google's largest products - AI Mode and AI Overviews in Search, Workspace features, and Android/Pixel - reinforcing Alphabet's core advertising and device businesses rather than billing directly.
Model licensing & partnerships (incl. Isomorphic)
Platform partnerships such as powering Apple's next-generation Siri, plus pharma collaborations run through the Isomorphic Labs spinout (Eli Lilly and Novartis deals valued near $3B) that apply DeepMind science to drug discovery.
07
Track record
How past announced dates turned out.
1/1announced date hit on time
1 Hit
0 Slipped
0 Missed
Hit
Fly Suncatcher TPU prototype on SpaceX Transporter-18
Target October 1, 2026 currently scheduled (Space.com); Google said next week from Sept 24 · Resolved Oct 1, 2026
SpaceX Transporter-18 launched 1 Oct 2026. Google Research confirmed the Planet-built Suncatcher prototype reached orbit, made contact, and was operating as expected. Detailed Trillium TPU health remains a later in-orbit data collection.
Google Cloud launches Gemini agent at Gemini at Work
On Oct. 8, 2026 Google Cloud announced the Gemini agent at Gemini at Work 2026 as a universal agent for work that plans tasks, uses skills and tools, connects to business systems, and chooses the best model. Googles consumer blog dated the launch Oct. 8. TechCrunch, reporting the event, said Sundar Pichai put Gemini at over 1 billion monthly active users and nearly 90 percent of Fortune 100 businesses on Gemini Enterprise; the agent starts with businesses before consumers; users can pick models including Anthropic Claude, with open-source and private models later. Early testers included On, Shopify, and PayPal. Googles Sept. 30 Argon post still rolls first to Fairwind cyber defenders, then paid API customers and Google AI Ultra, with no public date. This is an enterprise agent surface, not a public Gemini 4 Argon API.
Opens SynthID Detector to the public
On Oct. 7, 2026 Google DeepMind made SynthID Detector available globally in English for images, video, and audio from Google and named partners including OpenAI, Nvidia, and Kakao, with Apple listed as coming soon. Google said it has watermarked more than 180 billion images and videos and 240,000 years of audio.
Suncatcher prototype confirms contact after Transporter-18
On 1 Oct 2026 Google Research said the Planet-built Project Suncatcher prototype launched on SpaceX Transporter-18, reached orbit, made contact, and is operating as expected. Beals said the team will gather in-orbit TPU data on launch stress, radiation and thermal extremes over the coming weeks. This is satellite-level contact, not a published Trillium health readout.
Publishes SynthID Bio protein watermarks in Nature
On Sept. 30, 2026 Google DeepMind introduced SynthID Bio in a Nature paper and company blog as a proof of concept for watermarking AI-generated protein sequences and AlphaFold 3 predicted structures while preserving function. Wet-lab tests across VEGF-A, SARS-CoV-2 spike RBD and PD-L1 binders matched unwatermarked hit rate, binding affinity and sequence diversity. Nature news notes the watermark can be removed by redesign; DeepMind lists tamper resistance as a remaining challenge and is open-sourcing code and in vitro data.
Suncatcher TPU pathfinder launches on Transporter-18
Google's first Project Suncatcher prototype, carrying TPUs on a Planet Labs satellite, launched Oct. 1, 2026 on SpaceX Transporter-18 from Vandenberg. Google's Sept. 24 briefing said the mission will collect radiation, vibration and thermal data before a 2027 two-satellite optical-link test. Independent and Space.com confirmed liftoff; on-orbit TPU results have not been published.
Gemini 4 Argon rolls out to Fairwind defenders
On Sept. 30, 2026 Google DeepMind SVP Koray Kavukcuoglu announced Gemini 4 Argon, first available to trusted cyber defenders through the Fairwind Program. Google set introductory pricing at $2 per million input tokens and $10 per million output tokens, raised the output limit to 1 million tokens from 64,000, and reported 77.9% on DeepSWE v1.1. Paid API and Google AI Ultra access is next with no date. Fairwind and internal teams get a version without cyber guardrails; CWE-bench v1 was reported as a 68% tie.
Pichai signs White House Super Intelligence accord
On September 29, 2026 Alphabet and Google CEO Sundar Pichai signed the White House Accord on Super Intelligence Joint Commitment on Frontier Responsibilities. Pichai is Alphabet/Google CEO, not DeepMind CEO. The accord is a voluntary four-layer safety plan; Trump signed a same-day executive order directing federal agencies to use Super Intelligence and SI in place of Artificial Intelligence and AI.
Pichai attends White House AI lunch; America.gov uses Gemini
On September 29, 2026 Alphabet and Google CEO Sundar Pichai is attending a White House East Room lunch with President Trump and Speaker Johnson, company spokespeople told CBS News and Alphabet confirmed to CNBC. The same day Trump unveiled America.gov; a White House official told NBC News the portal uses Gemini and Grok.
2026
Kavukcuoglu says Gemini 4 is in post-training
On 24 Sep 2026 Google DeepMind chief Koray Kavukcuoglu told The Information, as reported by 9to5Google, that Gemini 4 had entered post-training, including safety testing, with internal use in Antigravity. He said Google intends to release an early post-training output as soon as possible and hopes that is much earlier than the end of 2026, without giving a date. He said the lab took a step back to focus on Flash models and that the current focus is Gemini 4; the same interview did not say whether Gemini 3.5 Pro will still ship. DeepMind's models page still labeled 3.5 Pro coming soon as of 29 Sep.
Ships Gemini 3.8 Live with Live Avatar in Gemini Enterprise
On September 24, 2026 Google introduced Gemini 3.8 Live with Live Avatar, pairing native live dialogue with near-real-time lip-synced video personas in Gemini Enterprise. Google said the feature supports 97-language switching, background tool calls, preset or allowlisted custom avatars, and SynthID watermarking on audio and video. The Verge confirmed it is enterprise-only. This is separate from the same-day Project Suncatcher Transporter-18 booking already on the timeline. Gemini 3.5 Pro remained labeled coming soon.
Google books Suncatcher TPU satellite on Transporter-18
On September 24, 2026 Google Research said Project Suncatcher's first prototype satellite, built with Planet, will fly on SpaceX's Transporter-18 rideshare to test Trillium TPUs in orbit. Google said ground radiation tests showed the chips can survive a dose greater than a five-year mission and that a 2027 follow-on will fly two satellites to test high-bandwidth laser links. Space.com reported Transporter-18 as currently scheduled for October 1; Google described the flight as coming next week.
Co-founder Shane Legg launches the DeepMind Institute
On September 16, 2026 Google DeepMind co-founder Shane Legg launched the DeepMind Institute to broaden public discussion of artificial general intelligence, spanning science, education, society, policy, philosophy and human flourishing, and opening with three essays on AI safety, economic policy and principles for a new utopianism. Legg told the Financial Times that capabilities must not get ahead of safety, called Dario Amodei's proposal to slow frontier releases interesting directionally and worth considering, said it is premature to declare AGI achieved despite recent claims by Nvidia and OpenAI executives, and reaffirmed his forecast of a 50 percent chance of minimal AGI by 2028.
Ships Gemini 3.8 Live and Live Extended Thinking
On September 15, 2026 Google DeepMind released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, native speech-to-speech models for real-time voice agents. Live is built for low-latency dialogue with visual grounding and 97-language switching; Extended Thinking adds background multi-step reasoning while the conversation continues. Both are available via the Gemini Live API and Google AI Studio, priced at $0.005 per minute of audio input and $0.018 per minute of audio output, and are rolling out in the Gemini app, Search and Workspace. Gemini 3.5 Pro remained labeled coming soon on DeepMind's models page the same week.
Releases AlphaGenome Atlas covering 9 billion DNA variants
On September 8, 2026 Google DeepMind launched AlphaGenome Atlas, a roughly 1-petabyte catalogue of predicted molecular effects for about 9 billion single-nucleotide variants - every possible single-letter change in the human genome - more than 30 times larger than the AlphaFold Database. The atlas includes an AlphaGenome Variant Impact (AVI) score combining AlphaGenome and AlphaMissense, plus motif maps, and is free for academic use via a web portal, the AlphaGenome API and a Google Antigravity skill, with commercial access planned on Google Cloud.
Launches WeatherNext 3 across Search, Gemini and Maps
On September 3, 2026 Google DeepMind and Google Research released WeatherNext 3, an AI weather model that ingests live geostationary satellite mosaics and station observations to produce hourly forecasts at up to 5 km resolution, about five times sharper than WeatherNext 2. DeepMind said precipitation scores improved by up to 60% versus NASA IMERG at early leads and that day-ahead rain forecasts in consumer products should be up to 50% more accurate, with new 100-meter wind and solar-radiation fields for clean-energy planning. The model is live in Search, Gemini, Maps, the Maps Platform Weather API, Earth Engine, BigQuery and Cloud Storage.
Releases Gemini 3.5 Transcribe for voice agents and dictation
On August 26, 2026 Google introduced Gemini 3.5 Transcribe, a speech-to-text model for real-time streaming and pre-recorded audio with speaker attribution, word-level timestamps and custom vocabulary. Google said it achieved a 4.0% average word error rate in streaming and 2.6% non-streaming on Artificial Analysis tests, covering 85+ languages, and is live in the Gemini API, Rambler on Android Gboard, Google Antigravity and the Gemini macOS app, with Chrome dictation still listed as coming soon.
Ships Gemini 3.8 Flash and launches Fairwind cyber-defense program
Google released Gemini 3.8 Flash on September 2, 2026, three weeks after Gemini 3.7 Flash, alongside a specialized Gemini 3.8 Flash Cyber variant for autonomous vulnerability discovery and automated patching - Google's Chrome Security team measured 2.6x more correct patches than larger commercial models, and its Cloud Vulnerability Research team used it to find a critical vulnerability in under two hours. Cyber access is gated through a new Fairwind Program giving vetted governments, healthcare providers, telecoms and critical-infrastructure operators (650+ partners including CrowdStrike, Palo Alto Networks and Snowflake) early access to the model paired with Google's CodeMender patching harness. The launch, read as Google pushing into Palantir's core government/defense AI territory, contributed to a 7% drop in Palantir shares the same day.
Pilots double-blind evaluations of a Gemini Flash Lite model
On August 27, 2026 Google DeepMind said it is running what it calls the first double-blind evaluation of a proprietary frontier-class model, testing a Gemini Flash Lite system against confidential benchmarks inside a cryptographically sealed Google Cloud Confidential Space environment. Partners include the Singapore AI Safety Institute, OpenMined, AVERI, and MLCommons; the setup is designed so evaluators cannot see Gemini weights and Google cannot see the test prompts, addressing benchmark contamination as models get more capable.
AI responsibility team moves out of DeepMind into Global Affairs
Reports on August 27, 2026 describe Google shifting a roughly 90-person AI responsibility team - covering chemical, biological, radiological and nuclear model risk, human-AI interaction, and the psychological effects of chatbots - out of Google DeepMind and into Google's Global Affairs organization (which oversees lobbying and public policy), effective early September 2026. Some DeepMind researchers voiced concern about losing independence and direct access to Gemini developers; VP Helen King, who leads the team, told staff its research mission, compute, staffing, and ability to publish externally would remain unchanged. The move follows Demis Hassabis's August 5, 2026 shift to Chair as Alphabet continues integrating DeepMind more tightly into Google's corporate structure.
Coding research team relocates from London to Mountain View
Reports in early-to-mid August 2026 describe Google DeepMind shifting AI leadership and its coding-focused research team, led by Sebastian Borgeaud, from its London headquarters to Mountain View, California - part of a broader consolidation of Alphabet's AI leadership in the Bay Area as the company races against OpenAI and Anthropic. The move follows the August 5, 2026 Hassabis/Kavukcuoglu leadership transition and comes amid a stretch of high-profile researcher departures.
Gemini 3.7 Flash ships as the new default workhorse model
DeepMind releases Gemini 3.7 Flash on August 13-14, 2026, succeeding Gemini 3.6 Flash as the default model for coding and agentic tasks just 23 days after its predecessor. It lifts the DeepSWE v1.1 coding benchmark from 49.0% to 65.3% and raises output speed from roughly 225 to 340 tokens per second, while Gemini 3.5 Pro remains delayed with no new release date given - Google has not disclosed a fix timeline for the coding-performance and reliability shortfalls that pushed the flagship model past multiple 2026 targets.
Gemini app crosses 1 billion monthly active users
Google discloses on August 11, 2026 that the Gemini app has surpassed 1 billion monthly active users, up from 900M+ reported at Google I/O in May 2026 and 950M+ at Alphabet's Q2 2026 earnings call in July - a roughly 11% jump in about three weeks that underscores the app's rapid growth since its rebrand from Bard.
EU orders Android/Search interoperability for rival AI assistants
On July 16, 2026 the European Commission adopted two binding Digital Markets Act decisions ordering Google to give rival AI assistants the same Android system access Gemini holds (invocation, context, actions, resources) and to share anonymized Search ranking and click data with rival search and AI services. Deadlines run from January 2027 (search-data sharing) to July 2027 (Android interoperability); non-compliance risks fines of up to 10% of Alphabet's global annual turnover. This is separate from the EU's ~$4.1B Android antitrust fine and bears directly on how Gemini is distributed.
Hassabis steps down as CEO; Jeff Dean departs Google
On August 5, 2026 Demis Hassabis moved from CEO of Google DeepMind to Chair of Google DeepMind and Chief Scientist of Alphabet, handing day-to-day operations and the Gemini roadmap to Koray Kavukcuoglu, an SVP now reporting directly to Sundar Pichai. The same day, longtime Google chief scientist Jeff Dean announced he is leaving after 27 years to co-found an AI-for-science startup, Discovery Loop, alongside fellow departing Googlers Sanjay Ghemawat, Oriol Vinyals and Quoc Le; Alphabet shares fell about 4% on the news.
Q2 2026 earnings: Gemini hits 950M MAU as AI capex surges past guidance
On July 22, 2026 Alphabet reported Q2 2026 revenue of $119.8B (+24% YoY), with Google Cloud up 82% YoY to $24.8B and a $514.0B backlog. Sundar Pichai disclosed the Gemini app had reached 950 million monthly active users (daily actives tripling over the prior year) and that model APIs were processing 22 billion tokens per minute, up from 16 billion the prior quarter. Alphabet raised its full-year 2026 capex guidance to $195-205B (from $180-190B) after $44.9B in Q2 capex alone, pushing free cash flow negative for the first time in the company's history as a public company.
EU's top court upholds ~$4.1B Android antitrust fine
On July 2, 2026 the EU's top court upheld a roughly 4.1 billion euro (~$4.67B) antitrust fine against Google over Android practices, closing out one of Brussels's longest-running cases against the company even as separate EU and US proceedings continue to bear on how Google can bundle and distribute Gemini.
AlphaFold team dismantled as DeepMind pivots to Gemini
Reports around July 29, 2026 said DeepMind has dismantled the dedicated team behind its Nobel Prize-winning AlphaFold system, having reassigned most of the original paper's authors over the prior year to Gemini, enzyme design, fusion, genomics and Isomorphic Labs work. The restructuring reflects a broader shift away from single-purpose research teams toward a Gemini-centered, agent-based structure; nearly a quarter of the original AlphaFold paper's full-time DeepMind authors have left the company entirely, including 2024 Nobel laureate John Jumper and co-authors Jonas Adler and Alexander Pritzel, who all joined Anthropic.
Gemini 3.6 Flash ships as the new default model
On July 21, 2026 DeepMind released Gemini 3.6 Flash - the new default workhorse model succeeding 3.5 Flash - alongside a Gemini 3.5 Flash-Lite tier and a Flash Cyber variant. It keeps a 1-million-token context window, advances its knowledge cutoff to March 2026, and uses roughly 17% fewer output tokens while scoring higher on coding, long-context and computer-use benchmarks, at $1.50/$7.50 per million input/output tokens. Google said Gemini 3.5 Pro remained in partner testing and that DeepMind had begun its most ambitious pre-training run yet for the next-generation Gemini 4.
Hassabis calls for a global frontier-AI watchdog
On July 14, 2026 Demis Hassabis publishes a manifesto, 'A Framework for Frontier AI and the Dawning of a New Age,' urging a U.S.-led, FINRA-style standards body to vet the world's most advanced models and coordinate an industry slowdown if dangers mount - a proposal he wants operational before year-end.
Talent exodus and Gemini 3.5 Pro delays
In June 2026 DeepMind loses star researchers - Nobel laureate John Jumper departs for Anthropic and Noam Shazeer is reported leaving - while the flagship Gemini 3.5 Pro slips past multiple launch targets (widely reported for July 17) after a ground-up architectural rebuild, raising questions about the lab's pace versus OpenAI and Anthropic.
Gemini Omni and 3.5 Flash unveiled at I/O
At Google I/O 2026 Google reveals Gemini Omni - a native world-simulator model merging Veo and Genie that understands physics like gravity and fluid dynamics - and Gemini 3.5 Flash, a frontier-class model at under half the price of rivals. The Gemini app is disclosed at 900M+ monthly active users (up from ~400M a year earlier), with monthly token processing over 3.2 quadrillion.
Gemini 3.1 Pro ships
On February 19, 2026 DeepMind releases Gemini 3.1 Pro in preview, more than doubling its predecessor on abstract-reasoning benchmarks (77.1% on ARC-AGI-2), available across the Gemini app, AI Studio, Vertex AI and NotebookLM.
Alphabet crosses $4 trillion; Apple picks Gemini for Siri
Riding Gemini 3's momentum and Google Cloud growth, Alphabet's market cap tops $4 trillion on January 13, 2026. Apple selects Gemini to power the next generation of Siri, with 'More Personal Siri' features slated for a Spring 2026 rollout.
2025
Gemini 3 takes the frontier lead
On November 18, 2025 Google DeepMind releases Gemini 3 Pro (with a Deep Think mode), claiming state-of-the-art results across math, science, multimodal and agentic benchmarks - scoring 1501 Elo atop LMArena and 76.2% on SWE-bench Verified.
AlphaGenome, AlphaEvolve and Gemini 2.5
DeepMind ships Gemini 2.5 (March 2025), the AlphaEvolve evolutionary coding agent (May 2025) that optimizes Google's own data-center and chip workloads, and AlphaGenome - the first model to read 1M base pairs of DNA at single-letter resolution - advancing its scientific-AI portfolio.
2024
Nobel Prize in Chemistry for AlphaFold
Demis Hassabis and John Jumper are awarded half of the 2024 Nobel Prize in Chemistry 'for protein structure prediction' via AlphaFold2 - a rare Nobel recognition of an AI achievement.
2023
Gemini 1.0 launches
In December 2023 Google DeepMind releases Gemini, its natively multimodal model family (Ultra, Pro, Nano), positioned as the successor to PaLM and the foundation for Bard (later renamed the Gemini app).
Google Brain and DeepMind merge
In April 2023 Google combines its Google Brain division with DeepMind to form a single unit, Google DeepMind, with Demis Hassabis as CEO - consolidating Alphabet's AI research to accelerate the response to ChatGPT.
2021
Isomorphic Labs spun out
Alphabet launches Isomorphic Labs, an AI-first drug-discovery company spun out of DeepMind and led by Demis Hassabis, to apply AlphaFold-style models to pharmaceutical R&D.
2020
AlphaFold solves protein folding
At the CASP14 competition, AlphaFold 2 predicts protein 3D structures with accuracy rivaling experimental methods, effectively solving a 50-year grand challenge in biology and setting up an open database of predicted structures.
2016
AlphaGo defeats Lee Sedol
DeepMind's AlphaGo beats world champion Lee Sedol 4–1 at the ancient game of Go - a landmark long thought to be a decade away - demonstrating deep reinforcement learning at superhuman level and catapulting the lab to global prominence.
2014
Acquired by Google
Google acquires DeepMind on January 26, 2014 for a reported $400–650 million, in what was then Google's largest European acquisition, keeping the research team based in London.
2010
DeepMind founded in London
Demis Hassabis, Shane Legg and Mustafa Suleyman found DeepMind Technologies in London (officially launched November 15, 2010) with the mission to 'solve intelligence, and then use that to solve everything else.'
09
Funding
$2.7Bdisclosed across 4 rounds with a stated amount
Cumulative disclosed raise · dated rounds
RoundAmountInvestors
Google acquisition
Jan 2014
~$400–650M (reported)
Google (now Alphabet)
Alphabet internal funding (ongoing)
2014–present
Funded via Alphabet R&D; parent capex guided to $180–190B for 2026
