Google DeepMind Releases WeatherNext 3 AI Weather Model

Google DeepMind and Google Research released WeatherNext 3, an artificial intelligence model for global weather forecasting that incorporates real-time satellite data. The model increases parameters 2.4 times over its predecessor, enabling hourly forecasts with a spatial resolution of 5 kilometers. Google reports that precipitation forecasts are 50% to 60% more accurate than WeatherNext 2.

By utilizing raw observations rather than relying on traditional six-hourly government data cycles, the model reduces the age of information in typical forecasts from roughly seven hours to three or four. Google claims WeatherNext 3 outperforms AI systems from Microsoft and Nvidia, as well as traditional physics-based forecasts from the U.S. National Weather Service. However, the startup WindBorne disputes Google's claim of being the first to use raw observations, asserting its WeatherMesh 6 model has done so since late 2025.

The model is being integrated into Google Search, Maps, and Gemini, and is available via Google Cloud. It specifically targets renewable energy needs by providing data such as wind speed at turbine height. The nonprofit Open Climate Fix has tested the model to help utilities and energy traders manage the rapid fluctuations of wind and solar generation in short-term markets.