Google DeepMind launches WeatherNext 3

Google DeepMind and Google Research introduced WeatherNext 3 on 3 Sep 2026 as their most accurate global weather model to date. It is a 64-member Functional Generative Network mesh transformer that takes live geostationary satellite mosaics plus ECMWF HRES analysis as direct input, initialises every hour (24 runs a day) and outputs up to 0.05° (~5 km) at station level, 0.1° gridded surface and 0.25° pressure levels, with a 15-day horizon on the 6-hourly cycles and 48 hours on the interim hourly runs. Google says precipitation forecasts a day or more out are up to 50% more accurate, with the biggest gains where forecasts were historically weakest; TechCrunch reports rain evaluations 60% better than WeatherNext 2 from a model with 2.4× more parameters, and that it tops Brightband's independent Operational WeatherBench, ahead of deep-learning models from Google, Microsoft, Nvidia and ECMWF as well as traditional NWS and ECMWF forecasts. It starts powering weather in Google Search, the Gemini app, Google Maps, the Maps Platform Weather API and Earth Engine today, with data for developers via BigQuery, Earth Engine and Cloud Storage.