Search “Google DeepMind”

Elorian Releases DecoyCount Benchmark: Frontier Models All Score Below 50% on Visual Counting Accuracy

Elorian and Mercor jointly released the DecoyCount benchmark, which tests whether models truly count objects or rely on visual shortcuts. Frontier models all scored below 50% accuracy, with the highest being GPT-6 Astra at 43.17%. Elorian says its visual reasoning model Glimpse scored 50.48% in a single run on 1,660 questions, making it the only model to exceed 50%, but this result comes from internal testing and did not use an agent or tool framework, whereas the six models in the paper each took the average of three runs. Elorian CEO Andrew Dai said the benchmark examines both vision and reasoning, and can distinguish whether a model failed to see the objects or miscounted them.

Google releases Gemini 4 Carbon, tops the Artificial Analysis leaderboard with 59 points

Google released Gemini 4 Carbon, which scored 59 points on the Artificial Analysis Intelligence Index v5.0 leaderboard, surpassing Claude Opus 5.5's 58 points and GPT-6 Astra's 57 points. The model achieved 88.2%, 54.8%, and 71.0% on SEC-bench Pro, ExploitGym, and CWE-bench, respectively, and Google says it can handle engineering problems in large codebases continuously for several days. The model this week is first being made available to some cybersecurity defenders, then expanded to trusted partners through the Fairwind Program, with no timeline yet for general users; initial pricing is $2 per million input tokens and $10 per million output tokens.

Google releases Gemini education research results in the UK and Italy

Google announced the results of its Gemini classroom research in the UK and Italy: a 10-week randomized controlled trial in the UK covering more than 1,100 students aged 12 to 15 found that AI tutoring performed similarly to human tutoring on platform tasks and quizzes, but showed no improvement on the STAR standardized test. A study of 31 schools and 1,600 teachers in Italy showed that after teachers used Gemini to co-design lessons, traditional lecturing decreased by 38%, student-led collaborative learning nearly tripled, and student engagement rose from 77% to 84%.

Physical Intelligence in Talks for at Least $1 Billion in New Funding, Valuation Could Reach $20 Billion

According to Bloomberg, U.S. robotics AI company Physical Intelligence is in talks for a new funding round of at least $1 billion, with existing investors Avenir and Lux Capital considering leading the round, and a post-money valuation expected at about $20 billion. The company was founded in 2024 and was valued at $5.6 billion in its November funding round last year; it currently has no commercial revenue and has not mass-produced robot products. Its technical approach is to build on VLM by introducing action generation capabilities to train VLA models, and it has iterated through π0, π0.5, and π0.7; this month it launched a beta version of a robot foundation model API, offering model fine-tuning and inference services to external developers.

Google previews Genie 4 world model, which can generate 3D environments with audio

Google previews the Genie 4 world model, which can generate realistic three-dimensional environments based on natural language descriptions and supports editing operations such as changing the weather and adding objects with follow-up prompts. Compared with the previous generation, Genie 4 is better at rendering complex details such as fluid motion, can also generate accompanying audio such as crashing waves, and makes virtual characters move naturally and be controllable through real-time interaction. Internally, the model calls Gemini Omni to ensure visual realism and Gemini 3.8 Flash to maintain inter-frame consistency. It is currently being tested with a small number of early users and will later be offered through the Google AI Ultra subscription, the Gemini API, and Google AI Studio.

Riley Goodside Says Prompt Engineering Is Dead: Ordinary Users Can Just Ask Directly

Riley Goodside, once regarded as the world's first officially titled prompt engineer, has publicly declared that "prompt engineering is dead"—ordinary users can ask whatever they want, however they want. He added that the workload of engineers who write underlying prompts for systems, fine-tune small models, and embed prompts into applications has not decreased; what has died is only the prompt-crafting skill that ordinary users were once required to have. Chainguard used a single sentence—"find remotely exploitable high-severity vulnerabilities in PHP"—to get its self-built system to reproduce known vulnerabilities and verify official fixes.

AI-driven cyberattacks erupt intensively; industry insiders call October a "Black October" for cybersecurity

Cailianshe reported that in recent weeks hackers have massively deployed AI agents and automated tools to launch attacks, and industry insiders describe this month as Black October in the cybersecurity world. A major vulnerability was exposed in PHP, and as many as 70% of websites worldwide may face the risk of being hacked. The U.S. National Security Agency and the FBI warned enterprises to shut down their websites if they cannot patch immediately, and some government websites in Canada, Belgium, France and the Netherlands have proactively gone offline. Shane Fry, chief technology officer of security firm RunSafe Security, said AI can weaponize a vulnerability into attack code within hours, whereas in the past this took weeks or even months.

What exactly are automakers aiming for by building humanoid robots?

Automakers building humanoid robots aim to incubate new businesses using the funding, real factory tasks, and supply chain capabilities of their automotive operations—first validating them in manufacturing and store scenarios, then pursuing external customers. Tesla's Optimus has already begun production at the Fremont factory, with external delivery targets still around the end of 2027, and plans to build 10,000 to 20,000 units for training; Figure 02 has accumulated more than 1,250 hours of operation at BMW's factory, and a forearm failure prompted a wrist structure change in the third generation. Hyundai controls Boston Dynamics, XPeng develops IRON in-house, and Mojia leverages Chery's channels—the approaches differ, but whether internal trials can translate into sustained paying external customers remains to be verified.

Global Embodied Intelligence Industry Development Analysis Report 2026

In the first half of 2026, global humanoid robot shipments were about 19,000 units, with Chinese manufacturers' share rising to over 97%; in 2025, global shipments were about 25,000 units, with China accounting for about 52%, and global embodied intelligence financing exceeded USD 11 billion. The report notes that the industry has entered its first year of mass production and a commercialization inflection point, but the ratio of intended orders to actual deliveries is only about 10%–20%, and commercialization remains a key bottleneck. China and the United States are positioned in a staggered manner along the value chain: the United States is strong in the "brain" and computing capital, while China is strong in the "body" and supply chain mass production.

Pichai Reviews Gemini 4 Argon: Why 3.5 Pro Was Cut

Google CEO Pichai revealed in the "Sources" interview that Gemini 4 Argon processed about 25 trillion tokens in 72 hours of launch, 3 times that of Gemini 3.0 in the same period, and is still rate-limited due to compute constraints. He explained for the first time the reason for cutting Gemini 3.5 Pro: the team found a gap in the model on Agentic Coding, first used the Flash version to complete intensive iterations from 3.5 to 3.8 in about 3 months, while Argon pretraining had already been completed at that time, so they injected the accumulated gains into Argon post-training. He also mentioned that Google internally has entered the RSI stage of researchers and models collaborating, and said the next-generation Gemini 5.0 will be much stronger than Argon.

PyTorch Conference 2026 held in San Jose, announces progress on TorchTPU and TorchNeuron

PyTorch Conference North America 2026 was held on October 20–21 in San Jose, attracting more than 3,500 attendees and focusing on accelerating open intelligence and projects including PyTorch, vLLM, DeepSpeed, Ray, Helion, and Safetensors. The event announced TorchTPU, a native PyTorch TPU backend developed in collaboration between Google and Meta, as well as TorchNeuron, a native backend for AWS Trainium, which entered public testing under Apache 2.0 on October 20. Shopify shared a case study of building a continuous learning loop with PyTorch and vLLM, saying it compresses production failures into model weights every day and has reduced serving costs by 96%.

Swiss robotics startup Enact Intelligence raises $30 million in pre-seed funding

Swiss robotics startup Enact Intelligence (Eta Robotics AG) has completed a $30 million pre-seed funding round, with investors including Yann LeCun, Mistral AI CEO Arthur Mensch, and Hugging Face co-founder Thomas Wolf. The company was founded by Manuel Yves Galliker, a former engineer at Physical Intelligence and 1X Technologies, and is developing an AI control model that can adapt to multiple robot embodiments, along with in-house actuators that combine force and tactile sensing capabilities. The funds will be used to hire engineers and upgrade technology.

Hugging Face Releases Open-Source DNA Model Carbon-X, Runs Locally on a Laptop

Hugging Face on Thursday released the open-source DNA understanding model Carbon-X, which can run locally on a laptop, letting users ask in natural language how DNA changes affect gene activity and health. The model was developed in collaboration with researchers at Zhongguancun Academy and the University of Naples Federico II; it extends the vocabulary of an existing language model to read DNA in six-letter units and was further trained on 300 billion tokens of DNA sequences, English text, and samples pairing DNA with biological descriptions. The model has 14B parameters and can process sequences of about 50,000 base pairs at a time; it cannot upload a complete human genome. The developers also note that the prediction accuracy has not yet been validated through wet-lab experiments.

Idealab founder Bill Gross opens free access to 1.2 million AI-generated business ideas

Idealab founder and CEO Bill Gross has launched idealab.ai, offering free access to 1.2 million AI-generated business ideas. These ideas were generated by combining 31 ideation methodologies with roughly 17 million source documents, including news articles, research reports, expired patents, public company filings, and community discussions. Users can search by industry, skill, interest, and region, and can also upload a résumé or LinkedIn profile to receive personalized ideas. Gross believes the surge in entrepreneurship brought about by AI can partially offset its impact on employment.

Eka Robotics raises $150 million at a $14.5 billion valuation to build robotic hands with a sense of touch

Eka Robotics has completed a $150 million Series A funding round at a $14.5 billion valuation, led by Andreessen Horowitz and Fidelity's F-Prime. Its vision-force-action model uses force sensors in custom grippers to sense grip strength, allowing it to adjust its hold mid-grasp when an object slips. The robot can pick up about 1,500 objects per hour that are outside its training data, compared with a human baseline of about 600. The gripper costs less than $1,000 and is currently rated for 5 million work cycles; the company aims to raise that to 20 million, the standard for industrial robotic arms.

a16z Leads Eka Robotics Series A Funding

a16z announced it has led the Series A funding round of Eka Robotics, betting on its approach to solving the challenge of robotic dexterous manipulation. Eka builds a Vision-Force-Action model that combines visual perception with robotic hand force feedback, and trains manipulation policies with reinforcement learning in physical simulation, enabling robots to obtain information from contact itself and correct in real time when slipping or insertion is obstructed. The founding team includes Pulkit Agrawal of MIT CSAIL and Tuomas Haarnoja, co-proposer of Soft Actor-Critic.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Elorian Releases Its First Model Glimpse, Using a Visual Draft Space for Image Reasoning

Elorian has released its first model, Glimpse, opening the API as a research preview, with a waitlist for the chat interface. During reasoning, the model can crop image regions, extract video frames, or run code to measure and compare what it sees, which Elorian calls a visual draft space. Glimpse is priced at $0.60 per million input tokens and $3 per million output tokens, and according to Elorian's evaluations it surpasses GPT-6 Astra on scientific chart reasoning and short video description. The company says the model was trained on several hundred GPUs, a scale notably smaller than the frontier systems it benchmarks against, and it initially targets enterprises with large amounts of video, such as robotics, research laboratories, and media and creative firms.