Google DeepMind

To be genuinely useful in human spaces, robots need the fine motor control we take for granted. 🤖 Gemini Robotics 2 enables diverse robotic platforms to master delicate manipulation - from packing and sealing a flexible Ziploc bag, to unscrewing a lightbulb, and using multi-fingered hands to tie off a trash bag. ↓

Gemini 3.7 Flash is here. It’s stronger for coding, knowledge work, and web development. 🔵 It shows strong gains over 3.6 Flash in key coding tasks like debugging and issue resolution. 🔵 It’s better at designing more functional web layouts and apps using fewer prompts. 🔵 It delivers improved reasoning and accuracy when completing real-world business workflows. Developers can try it now in Google Antigravity, with API access in Google AI Studio and Android Studio. Google AI Pro and Ultra subscribers can use 3.7 Flash in Gemini Spark in the Gemini app. Find out more → https://goo.gle/4g7mpaK

We’re launching SL2T: our new sign language-to-text model - and it’s coming to Android. Starting with American Sign Language-to-English on Pixel 11, people can sign directly into Gboard and Live Transcribe instead of typing. The system is the most capable sign language translation model ever, and optimized for real-world usability, such as one-handed signing. We built SL2T with the Deaf community - guided by Deaf Googlers, and our AI Sign Language Advisory Committee. Bringing ASL input to phones is just the beginning and we’re working to expand this technology to more languages and devices soon → https://goo.gle/45WnOfy

Packing for a sports game sounds simple to a human - but for a humanoid robot, it requires hundreds of physical decisions made in fractions of a second. 🤖 Watch Apptronik’s Apollo 2 use Gemini Robotics 2 to process a command, check a family calendar, and retrieve the right gear.

Meet Apollo 2 👋 Powered by Gemini Robotics 2, it's learning to master tasks that humans find simple, but robots find complex - like tying a trash bag or packing a gym bag. From solo chores to working alongside Duo, we’re one step closer to physical AI that’s truly helpful in the real world.

A single day of extra warning can help save lives when preparing for severe storms. Published today in Nature Magazine, WeatherNext is our breakthrough AI model unifying cyclone track, intensity, and wind structure forecasting. By providing a 24+ hour lead-time advantage over operational systems on average, it delivers a decade of progress in a single leap. During the 2025 season, it helped forecasters predict Hurricane Melissa’s rapid intensification and landfall as a Category 5 storm. Developed with partners at National Hurricane Center, Cooperative Institute for Research in the Atmosphere, UK Met Office, and Google Research, we’re releasing the code and weights of WeatherNext 2, as well as a new WeatherNext 2-mini to run in a free Google Colab notebook. Explore the research → https://goo.gle/4gdgJ02

🧬 What are the highest-priority "learnable" problems in antimicrobial resistance, and what data would it take to solve them? Today The Align Foundation, with support from Google DeepMind, are sharing Roadmapping the Future of AI-Enabled AMR Research: A Landscape of Community Priorities (https://lnkd.in/eewYx4ct). AMR was associated with nearly 5 million deaths in 2019, and could cause up to 10 million a year by 2050 if we don't act. It's also full of the kind of problems AI is well suited to: complex, multimodal, but held back by data that's fragmented and hard to compare across labs. To find where AI could actually help, we asked the community. We interviewed 130+ people across 20+ countries, convened 60+ experts in six workshops, and prioritized 10 problem areas where open, AI-ready datasets could move things forward. Working groups are now turning those priorities into dataset and model proposals. This answers the first call to action from the AI for AMR report, from Google DeepMind and the Fleming Initiative: build consensus on the problems AI could address. It's a starting point rather than a finish line, and the technical proposals come next, stay tuned 👀 . #AMR #AntimicrobialResistance #OpenScience #BioAI #MachineLearning #GlobalHealth #DreamDatasets

One brain. For any robot. 🤖 Meet Gemini Robotics 2: our next-generation AI model bringing intelligent whole body control, advanced dexterity, and multi-robot collaboration to embodiments of different shapes and sizes. Three new models drive this breakthrough: 1️⃣ Gemini Robotics 2: A vision, language, and action model controlling full humanoids from feet to fingertips. 2️⃣ ER 2: A reasoning agent built to plan complex, end-to-end tasks lasting several minutes. 3️⃣ On-Device 2: Runs locally and adapts to new robot bodies in just a few hours. Gemini Robotics 2 scales dexterity across different embodiments - from helping a five-fingered, 22-degree-of-freedom hand tie knots to performing high-precision kitting with two-fingered grippers. It even powers multi-robot teamwork so different machines can share the workload. Explore the models → https://goo.gle/4fIChAj

There is a lot of talk about how AI is transforming the way we work and live. If you strip away the noise, how are people actually using these tools right now? This week, Google published its first iteration of the AI & Economy ATLAS, an ongoing, large-scale study of how people are using Google’s AI products across 150 countries and 800 occupations. To make sense of this quickly moving landscape, the team used Google DeepMind’s Observation Clustering and Taxonomy Organisation (OCTO). OCTO allowed us to take massive, unstructured text data and distill it into clear, organised trends, all while applying the strictest privacy protections and removing sensitive information. While this is just a snapshot of a moment in time, here are a few things we're learning that challenge some common assumptions: - AI use at work is broad, but shallow: People are using AI selectively. In a typical job, AI is used for only about 21% of tasks. - It's a blue-collar tool, too: We are seeing industrial mechanics and tradespeople use conversational AI as a live collaborator to interpret complex test results or debug wiring. - The "life admin" assistant: A massive amount of AI use (over 86%) happens outside the office, helping people tackle household friction like researching purchases or licensing. Understanding how people actually interact with AI in the real world is one of the most critical steps in deploying it responsibly - and here we have possibly the world's most comprehensive study on AI and the economy. This is also just a starting point, things are evolving quickly and our aim is to share what we’re seeing with the world - take a look: https://lnkd.in/gxVCjPAp