Search “Google DeepMind”

English translation is being prepared.

English translation is being prepared.

Google DeepMind releases Gemini 3.7 Flash

On August 13, 2026, Google DeepMind released Gemini 3.7 Flash, calling it its strongest flagship model for coding and agents, just three weeks after the release of Gemini 3.6 Flash. Official figures state that the model scored 43.6% on FrontierCode 1.1 Main (34.4% for 3.6 Flash), 65.3% on DeepSWE v1.1 (49.0%), 1588 versus 1538 on WebDev Arena Elo, 34.0% versus 22.0% on GDP.pdf, and 30.4% versus 17.0% on AutomationBench; through the end of the year, introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens, half the original price of 3.6 Flash, and it comes with updated safeguards against CBRN and cyberattack misuse. The next day, Google AI Developers showcased it building a 3D web game with Three.js in Antigravity, as well as generating native code for Flutter, SwiftUI, Jetpack Compose, React Native, and NativeScript from a single architecture spec. On August 17, the company officially demonstrated it extracting historical plants from a Victorian botanical illustrated guide PDF and classifying them according to APG IV; on August 18 it showcased

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

Google releases the Gemini 4 Carbon model

On October 28, 2026, Google DeepMind released Gemini 4 Carbon, calling it its strongest model, ranking first with 59 points on the Artificial Analysis Intelligence Index v5.0, and scoring 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. The model targets long-horizon engineering tasks and was first opened to some cybersecurity defenders, then expanded through the Fairwind Program, and afterward made available to developers, enterprises, and consumers. On October 29, Arena.ai said Gemini 4 Carbon ranked second in Agent Arena with a net improvement of 16.2% and pushed the Pareto frontier forward; the same day, Yuchen Jin cited Artificial Analysis data to raise reservations, noting that the evaluation used the Antigravity CLI. On October 30, the Artificial Analysis Intelligence Index showed it in first place with 59 points, 1 point higher than Claude Opus 5.5. On October 31, Perplexity CEO Aravind Srinivas shared an early evaluation, saying it set a new SoTA on the WANDR benchmark.

Nathan Lambert criticized the Gemini model for publishing scores despite not being publicly available

On October 29, 2026, Nathan Lambert publicly criticized Google DeepMind's Gemini model: he stated that every time he wanted to try the model, it was not publicly accessible, and he believes that if the outside world cannot use it, there is no need to publish benchmark scores. Currently, this criticism appears only in an article he himself published, and there has been no response from Google DeepMind or other subsequent developments.

Google establishes AI research hub in Singapore

On October 29, 2026, Google announced the establishment of a Google Research hub in Singapore and a partnership with Singapore's Ministry of Health to deploy frontier AI into the country's healthcare system. The hub focuses on foundational AI models and also partners with the National University of Singapore and Nanyang Technological University to cultivate research talent. Google.org and Google DeepMind invested $7 million in the Philanthropy Asia Alliance's Health for Human Potential Coalition, of which $1 million is for the PathGen disease detection platform. Google DeepMind will also work with the National University Health System to evaluate AI clinical assistant tools, and Google Cloud is partnering with NHG Health on a population health project.

Google releases world model Genie 4

On October 27, 2026, Google DeepMind released the world model Genie 4, focused on real-time interactive simulation and introducing native audio to the series for the first time. According to the official introduction, the model is built on capabilities from the Gemini family: Gemini Omni is responsible for making the simulation match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain the simulation's coherence; users can control scenes in real time using natural language, and each world also instantly generates ambient sound, music, and interactive dialogue. The next day, Google DeepMind further explained that Gemini 3.8 Flash breaks down goals in a loop and checks frames to maintain causal coherence across multi-step events, and that Gemini 4 Argon has also been used to jointly simulate tasks with Gemini Robotics; the company said it remains an early experimental model, that object permanence and complex physical interactions still drift during long runs, that all generated content carries a SynthID watermark, and that it is currently being previewed to trusted testers first. On October 28, Ethan Mollick posted that Genie 4 can run DOOM, and tested it generating a game-like environment as it goes, saying the experience is not perfect. On October 29, Peter Yang used Genie 4 to build 5 worlds in real time, arguing that world models are the next big direction for AI after LLMs.

UK announces establishment of national AI biosecurity laboratory

The UK announced the establishment of the National AI Biosecurity Laboratory (AIBL). According to a report on October 24, 2026, the laboratory is positioned as an independent non-profit institution, with £100 million funded by the government, OpenAI and Anthropic jointly pledging up to £200 million, for an overall scale of up to £300 million. Initially it will focus on early detection of emerging biological threats and accelerated development of vaccines and treatments. OpenAI will provide £100 million in computing power and model credits, and open model access through the Rosalind Biodefense program; Anthropic will provide up to £100 million in model credits and technical support; Google DeepMind and Basecamp Research are participating in research collaboration. Subsequent reports likewise mentioned that the government is spending £100 million, OpenAI and Anthropic are providing model usage credits worth up to £200 million, and stated that Google DeepMind and Basecamp Research are participating in the collaboration. There are currently no further developments with evidence.

Hassabis shortens AGI timeline to two to three years

Google DeepMind founder and Nobel Chemistry laureate Demis Hassabis offered his judgment that artificial general intelligence (AGI) may arrive within the next two to three years. Earlier this year he had estimated three to five years. Hassabis defines AGI as a system that can match the full cognitive scope of the human brain, and uses the "Einstein test" as the criterion, namely whether a system trained only on knowledge from before 1901 could reproduce Einstein's 1905 papers. Other reports say that in an interview with The Wall Street Journal he shortened the timeline to the next 2 to 3 years, and mentioned that earlier expectations ranged from 5 to 10 years, to within 5 years, to 2030. There are currently no further developments confirmed by factual evidence.

Gray Swan launches revamped Arena red team testing platform

On October 20, 2026, Gray Swan launched a redesigned Arena red team testing platform. According to Fortune AI, contestants compete for prize money by tricking AI models into breaking through their guardrails; the new platform adds a sandbox testing environment for indirect prompt injection, and offers beginner-oriented challenges, training certificates, and a pathway for top performers to enter the Gray Swan red team community and take part in real testing. The report also states that Gray Swan conducts pre-release stress testing for frontier labs such as Anthropic, OpenAI, and Google DeepMind, and that its Arena community has about 15,000 members. There are currently no subsequent developments.

Metis Reasoning is founded in London, focusing on a compute-efficient AI approach

On October 20, 2026, the London-based AI company Metis Reasoning was officially established and launched, focusing on a compute-efficient reasoning approach as it seeks a direction distinct from the frontier models of Anthropic and OpenAI. According to Tech.eu, the company was co-founded by Thore Graepel, former head of research at Google DeepMind and co-creator of AlphaGo, and has a team of 12 people spanning researchers, mathematicians, robotics experts and technology entrepreneurs. The company believes that LLMs have fundamental limitations in reasoning, with results that are difficult to verify and lacking an audit trail; it says its approach is more verifiable, more robust and safer, and that it has already achieved some positive results on frontier reasoning problems at a cost far below that of frontier models. The report says its initial funding is believed to be in the tens of millions of dollars, but Metis has not yet announced a release date for its first AI model.

Arena.ai says user satisfaction is the biggest gap for AI labs

Arena.ai puts forward a judgment: whether users enjoy collaborating with agents is the biggest gap between AI labs. This judgment is based on Arena's tracking of four Agent Arena signals for the best agents from six labs—Anthropic, OpenAI, Google DeepMind, Kimi, xAI, and DeepSeek—from late June to early October. Among them, on the Praise vs Complaint metric, OpenAI and Anthropic were +35 points above the leaderboard average, while Moonshot, xAI, and DeepSeek ranged between +5 and +13; on Confirmed Success, the average lab's best model rose from -4 to +14 points within three months, with Anthropic leading at +20 with Claude Opus 5.5 (Max); the Bash Recovery gap narrowed from 34 to 16 points, and Steerability from 19 to 11 points. There are currently no further developments.

Andreessen Horowitz leads Eka Robotics Series A funding

On October 20, 2026, Andreessen Horowitz (a16z) led the Series A funding round of Eka Robotics. According to a TechCrunch AI report that day, Eka Robotics completed a $150 million Series A funding round led by a16z and F-Prime, at a post-money valuation of $1.45 billion; the company was founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, and its vision-force-action model, paired with force-sensor grippers, can grasp about 1,500 objects weighing under 1 kg per hour with a success rate exceeding 95%; the grippers cost less than $1,000 and are rated for 5 million cycles, and the company is currently conducting pilot tests of pick-and-place and cable manipulation with industrial customers. The next day, Zhidongxi reported likewise that this funding round was $150 million at a post-money valuation of about $1.45 billion, co-led by a16z and F-Prime with participation from Samsung Next and others, and said the company was founded in 2025 and is carrying out pilot cooperation with multiple industrial customers. There is currently no further progress following the completion of this funding round.

English translation is being prepared.

Four Gemini 4 Argon agents complete Minecraft without human intervention

The Google DeepMind team had four AI agents running on Gemini 4 Argon autonomously complete Minecraft in a randomly generated world, killing the Ender Dragon with no human input throughout. The team did not assign them roles, strategies, or game-specific tools; the four agents divided the work among themselves, warned each other of threats, recovered items from fallen teammates, and recorded mistakes in a shared memory. The entire run took 27 hours and 14 minutes, during which they wrote a total of 2,783 JavaScript programs, exchanged 229 chat messages, and died 605 times, costing $1,620 at the standard Gemini 4 Argon pricing that takes effect in 2027.

View story:Four Gemini 4 Argon agents complete Minecraft without human intervention

Chollet says Argon performs similarly to Astra on ARC-AGI-3

François Chollet commented that Gemini 4 Argon performs similarly to Astra on ARC-AGI-3, with Google returning to the frontier. On the same day, ARC Prize published the verified results for Gemini 4 Argon on the ARC-AGI series: ARC-AGI-3 scored 97.5% under the provider adapter harness at a cost of $4.8K; under the standard harness it scored 49.8% at a cost of $5.8K. ARC-AGI-2 scored 91.3% at $0.32 per task; ARC-AGI-1 scored 98.5% at $0.16 per task. ARC Prize said Google DeepMind is the second frontier AI lab to exceed 95% on ARC-AGI-3, and Argon's 97.5% is second only to GPT-6 Astra's 99.9%. The standard harness test could not complete all 55 games due to API issues; the high-reasoning setting completed only 19 games, so no score was published.

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.