Google DeepMind Releases WeatherNext 3, Its Most Accurate AI Weather Forecasting Model Yet
aiinsider ·
Google DeepMind and Google Research have unveiled WeatherNext 3, an AI weather forecasting model that will soon supply weather data to Google Search, Maps, and Gemini, while also being made available to researchers on Google Cloud platforms. The model has outperformed leading competitors such as Microsoft, Nvidia, and the European Center for Medium-Range Weather Forecasts on the Operational WeatherBench benchmark, as well as traditional national weather agency forecasts. WeatherNext 3 offers 5 km resolution on key variables, a 60% improvement in rain prediction over its predecessor, and generates hourly forecasts instead of standard six-hour intervals, aided by its ability to ingest real-time satellite data. According to DeepMind's Ferran Alet, machine learning forecasts weather by learning patterns from large data sets to approximate chaotic and noisy physics that traditional systems struggle to compute. Brightband's Daniel Rothenberg noted that the model's capability to target forecasts at specific weather stations brings AI predictions closer to real-world applications, like forecasting conditions at an airport. Google claims WeatherNext 3 is the first high-resolution global model to directly incorporate raw satellite observations, although rival WindBorne has used similar techniques since late 2025.
aiinsider