Google DeepMind and Google Research introduced WeatherNext 3 on September 3, 2026. The model uses real-time geostationary satellite inputs and refreshes hourly. Google says the system improves precipitation forecasts by up to about 50% when planning a day or more ahead in regions that have historically been hardest to forecast, citing Brightband evaluations. Surface fields include temperature and moisture at about 5 km, with other surface variables near 10 km and atmospheric/wind fields near 25 km. WeatherNext 3 powers weather experiences in Google Search, Gemini, Google Maps, the Maps Platform Weather API, and Earth Engine. Developers can pull hourly-ready data through BigQuery, Earth Engine, and Cloud Storage bulk download without self-hosting the model.
