Google's WeatherNext 3 forecasts hourly from live satellite data at five-kilometer resolution

Google Research and DeepMind have released WeatherNext 3, an AI weather model that learns directly from real-time geostationary satellite observations rather than from traditional physics simulations. Previous AI weather models, including WeatherNext 2, trained on numerical weather prediction data. Google says those simulations run on supercomputers and carry a six-hour delay, which introduces errors for fast-changing variables such as rainfall and temperature. WeatherNext 3 instead generates a fresh forecast every hour based on the latest satellite data, at a resolution of up to five kilometers. Google describes this as five times the detail of the predecessor model, allowing it to capture local weather events and terrain effects more precisely. The company says precipitation forecasts are up to 50 percent more accurate, and the model also produces outputs tailored to renewable energy applications. The forecasts are now live across Google Search, Google Maps, and the Gemini app. The shift matters because weather models that depend on numerical simulations inherit their latency, while a model that reads satellite feeds directly can react to rapidly developing conditions. Users of Google's consumer products receive the new forecasts automatically, and energy operators are among the groups Google is targeting with the specialized data.