The new AI weather model uses live satellite data to update forecasts every hour, with key temperature and humidity predictions available at around 5-kilometre resolution. Google DeepMind and Google Research have introduced WeatherNext 3, an AI-powered global weather forecasting model designed to deliver more frequent and localised predictions.
Unlike its predecessor, WeatherNext 2, which operated on a 25-kilometre grid and refreshed forecasts every six hours, WeatherNext 3 uses live geostationary satellite data to generate new forecasts every hour. Its temperature and humidity predictions are available at around 5-kilometre resolution. The model is aimed particularly at improving forecasts for rapidly changing weather conditions, including rain and snow. Google says WeatherNext 3 can provide up to 50% more accurate precipitation forecasts when planning a day or more in advance, while evaluations showed improvements of up to 60% against certain precipitation benchmarks. WeatherNext 3 also incorporates observations from weather stations to improve how its predictions represent conditions on the ground.
The technology is being integrated across Google Search, the Gemini app, Google Maps, Google Maps Platform and Google Cloud. Beyond consumer weather information, Google says the model can support sectors such as agriculture and renewable energy, where forecasts for variables including wind speeds, cloud cover and solar radiation can inform planning and operations. The launch comes as AI-based forecasting models increasingly complement traditional physics-based weather systems. WeatherNext 3 builds on Google’s earlier forecasting models while focusing on three areas—real-time observations, higher resolution and more frequent updates—to improve predictions for local and fast-changing weather conditions.
