WeatherNext 3

"Increasing resolution and performance of global weather models." DeepMind's third WeatherNext generation is a Functional Generative Network mesh transformer and the first in the line to ingest live hourly geostationary satellite mosaics alongside conventional analyses. It forecasts hourly at 5 km for surface temperature and moisture, 10 km for other surface variables, and 25 km aloft, against WeatherNext 2's 25 km and six-hour steps, a picture Google calls roughly five times sharper, and improves precipitation CRPS by 60% against IMERG, 30% against MRMS, and 10% against rain gauges at early lead times. It now powers weather in Search, the Gemini app, Maps, and the Maps Platform Weather API, with forecasts served through BigQuery, Earth Engine, and Cloud Storage. Weights are not released. Twenty-five authors led by Stephan Rasp and Boris Babenko; WeatherNext 2 is not tracked here.