Google DeepMind Launches WeatherNext 3 with Hourly Updates and Enhanced Accuracy

On September 3, 2026, Google DeepMind and Google Research introduced WeatherNext 3, a groundbreaking AI-powered global weather model that generates forecasts every hour. This innovation marks a significant departure from traditional weather forecasting methods, which typically rely on data updates every six hours.

The core advancement of WeatherNext 3 lies in its ability to utilize live geostationary satellite observations. By training directly on real-time data, the model can produce forecasts that offer a much more immediate and accurate representation of atmospheric conditions. With a spatial resolution of up to 5 kilometers for surface variables, WeatherNext 3 provides a level of detail that is approximately five times sharper than its predecessor, WeatherNext 2.

In terms of precipitation forecasting, WeatherNext 3 shows remarkable improvements, boasting up to 50% greater accuracy for rain and snow predictions over lead times of a day or longer. The model generates 64-member ensemble forecasts, with select cycles extending predictions up to 15 days ahead.

This latest model builds on a series of advancements by Google DeepMind in weather forecasting, starting with GraphCast in 2023, which demonstrated the potential of machine learning in medium-range predictions. WeatherNext 3 is particularly adept at predicting cyclone tracks and intensities, making it a valuable tool for sectors like renewable energy and agriculture, where precise weather data is crucial.

Google is integrating WeatherNext 3's outputs into various platforms, including Search and Maps, while also making the data accessible through BigQuery on Google Cloud. This integration aims to enhance decision-making processes across multiple industries that rely on accurate weather forecasting.