Introducing WeatherNext 3, our most advanced and accurate global weather AI model

WeatherNext 3 uses live satellite observations to refresh forecasts hourly and represent some surface conditions on a five-kilometer grid. Google DeepMind says its predecessor used a 25-kilometer grid with forecasts in six-hour increments. The announcement describes combining satellite inputs with training on weather-station observations to capture local conditions more closely.

The finer detail is particularly relevant where terrain or coastlines produce substantial differences over short distances. Resolution varies by variable: the company specifies five kilometers for temperature and moisture, ten for other surface variables and 25 for atmospheric variables. The update therefore provides more frequent, localized information without making every forecast field equally detailed.