Google launches WeatherNext 3, an AI weather model with five-times-sharper resolution

Google has released an updated version of its AI-based weather model, WeatherNext 3, promising more accurate forecasts, particularly for rain and snow. The new model produces a global forecast at what the company calls "unprecedented" resolution — five times sharper than the previous generation — by learning from real-time satellite observations.

The numerical jump is stark. The previous model, WeatherNext 2, generated a forecast every six hours on a 25-kilometre grid. The new version can render variables such as temperature and humidity at up to 5-kilometre resolution and refreshes the forecast hourly using the latest satellite observations. Google says the live data feed closes gaps in regions where ground-based rain gauges are sparse, mainly outside the United States and Europe, and that accuracy in predicting precipitation a day or more ahead has improved by up to 50 percent.

The extra speed and resolution are especially valuable for forecasting precipitation driven by fast-moving weather systems. Traditional physics-based models require solving complex equations on supercomputers, creating inherent latency; AI models spot patterns in historical data and produce forecasts faster. WeatherNext 3 takes this a step further by ingesting fresh satellite observations directly, rather than relying solely on historical training data.

The model was also designed to serve renewable-energy production — for example, wind speed at 100 metres, the typical hub height of modern turbines. "As the energy needs of Google and of humanity as a whole grow, making renewable energy an attractive option is a strategic interest for the company," said Ferran Alet, a researcher at Google DeepMind.

WeatherNext 3 is already integrated into Search, Maps, Gemini and other Google products. The company is collaborating with the US National Hurricane Center and agencies in Asia to improve forecasting. Google stresses, however, that the model is still trained on data from physics-based models, and that meteorological services continue to rely on a suite of forecasts before issuing official warnings.