Google AI Updates: September 4, 2026

1. WeatherNext 3 Trains on Live Satellite Data Instead of Simulation Output

Google DeepMind. WeatherNext 3 shifts training away from traditional numerical weather prediction simulations toward live global geostationary satellite data that updates hourly, plus sparse weather station observations that capture local topography. A Functional Generative Network mesh transformer natively emits dense gridded fields, discrete cyclone tracks, and station-level predictions from one model. Resolution is 5km for surface temperature and moisture, 10km for other surface variables, and 25km for atmospheric variables such as wind speed, roughly five times sharper than WeatherNext 2. Precipitation CRPS improves by up to 60% against IMERG, 30% against MRMS, and 10% against rain gauges at early lead times, with medium-range precipitation up to 50% more accurate. The model also emits 100m wind speeds for turbine-height forecasting and high-resolution cloud cover for solar planning. It is live across Search, Gemini, Google Maps, the Maps Platform Weather API, and Earth Engine, with developer access through BigQuery, Earth Engine, and Cloud Storage. Source