
Life update: Maria @BauzaVillalonga and I have graduated, moved to London and joined @DeepMind! We are looking forward to starting exciting research projects with fantastic teams.

Research Scientist @GoogleDeepMind: AI for Science & Sustainability PhD from @MIT_CSAIL BlueSky: bsky.app/profile/ferranalet.…

Life update: Maria @BauzaVillalonga and I have graduated, moved to London and joined @DeepMind! We are looking forward to starting exciting research projects with fantastic teams.

Today we released the datasets behind WeatherNext 3's precipitation gains! WN3 precip actually forecasts the output of another ML model, PARDIG! It's created by predicting precip from a high-quality, sparse satellite using lower-quality dense observations (SatWeave+analysis).
Accurate weather forecasting starts with how clearly we can observe the atmosphere in real-time. 🌍 That’s why we’re releasing PARDIG and SatWeave - the foundational datasets behind WeatherNext 3 - to help researchers and developers build new climate applications 🧵

We just presented Gemini Robotics 2.5 at CoRL 2026 in a live demo! Now our humanoid robot can successfully perform even more highly dexterous tasks—from using tools like tongs to tying a bow on a box entirely autonomously. 🦾🧠 deepmind.google/blog/gemini-…

@GoogleDeepmind Weather is hiring Student Researchers for next year! There are a few different projects. Candidates should have a strong ML background and interest in applying those skills to advance weather prediction. 1/3

Our work @GoogleDeepMind on AI cyclone forecasting is on today's @Nature cover! 🌪️ The art merges: - @NHC_Atlantic-inspired warning colors - Mesh & grid computations from the underlying weather model - Probabilistic "spaghetti" tracks Paper: nature.com/articles/s41586-0…

Today, @GoogleDeepMind and @GoogleResearch are introducing WeatherNext 3, our most advanced and accurate global weather AI model to date. It uses real-time satellite data to generate hourly high-resolution forecasts, precise precipitation forecasting, and clean energy variables. Most AI weather models are trained on data from numerical weather prediction (NWP) models, which come with a six-hour data lag. But by training on real-world observations, WeatherNext 3 is able to bypass these traditional constraints — meaning more people can get more accurate predictions to help them plan ahead. Starting today, WeatherNext 3 will power forecasts within Search, @GeminiApp, @GoogleMaps, Google Maps Platform Weather API and Google Earth Engine.

Today we are announcing WeatherNext 3! 🌦️ Better rain forecasts, higher spatial and temporal resolution, and a new state-of-the-art in medium range weather forecasting 🏆 Paper: storage.googleapis.com/deepm… Brightband's "Operational WeatherBench": owb.brightband.com/
Excited to share WeatherNext 3 from @GoogleDeepMind and @GoogleResearch! High-res temperature and precipitation forecasting using the latest satellite observations 🛰️ It's also the best global weather model according to Brightband's live leaderboard! owb.brightband.com/

We have also seen improvements on tropical cyclones on top of the 24 hour leap we recently published @NaturePortfolio We're partnering with @NHC_Atlantic and other agencies to help them safely deploy WN 3 to protect their communities. Check it out at deepmind.google.com/science/…

Excited to share WeatherNext 3 from @GoogleDeepMind and @GoogleResearch! High-res temperature and precipitation forecasting using the latest satellite observations 🛰️ It's also the best global weather model according to Brightband's live leaderboard! owb.brightband.com/


Our new Nature paper describing our WeatherNext cyclones model is out! A decade’s worth of progress in track and intensity forecasting. And it’s fully open source, Including a mini version which runs in a free colab (and is still really good!)
Very excited to share our @Nature paper today! 🌀 WeatherNext Cyclones predicts cyclone track, intensity & wind structure with 24+ extra hours of lead time over previous state-of-the-art, on average. This leap is equivalent to a decade of progress in the weather community🚀

Predicting cyclones accurately can help save lives - and every hour of lead time counts. Published in @Nature, our AI model WeatherNext achieves state-of-the-art accuracy in forecasting a storm’s track and intensity, giving us a critical extra 24 hours to prepare on average. 🧵

We made code, weights & a mini version you can run for free in Colab, all open source! 📄 nature.com/articles/s41586-0… 💻 github.com/google-deepmind/w… 📓 colab.research.google.com/gi… Thanks to fellow corresponding authors @tom_r_andersson, @IlanPrice, @stratosmarkou & the whole team!

The work merges two ideas: 1) Co-training end-to-end on global atmospheric data (20TB) + a tabular database of 5000 cyclones to model extremes 2) Functional Generative Networks: SotA weather forecasting via training on probabilistic marginals with stochastic weight perturbations

This work had real-world impact. ML experts @GoogleDeepMind @GoogleResearch collaborated with forecasters @NWSNHC @metoffice & CIRA to deploy the model. This contributed to a historic NHC forecast of Cat 5 intensification from just Cat 1 for Melissa. 📽️youtu.be/wd5ZZV8if54?si=OyGa…

Intensity forecasting has historically needed very high-res models (~2km grids). WeatherNext only requires 28km: 100x coarser! How the model achieves this shows there's more signal at coarse scales than previously believed, opening interesting questions for the community.