Google upgrades AI weather forecasting with WeatherNext 3

The new model achieves a five-fold increase in resolution compared to its predecessor, shifting from a 25-kilometer grid to a 5-kilometer grid. By processing live satellite observations every hour, the system bypasses the time-lag inherent in traditional physics-based supercomputer simulations. Research engineer Samier Merchant noted that the model moves beyond standard historical datasets by integrating fresher, high-fidelity observational inputs.

This technical leap provides a notable boost for detecting fast-moving weather systems. Google reports that precipitation forecasts are up to 50 percent more accurate when looking 24 hours ahead. Beyond standard weather tracking, the company is tailoring the model to support renewable energy, such as predicting wind speeds at the 100-meter height required for turbine efficiency. WeatherNext 3 is currently being integrated into Google Search, Maps, and Gemini, while the company continues to collaborate with the US National Hurricane Center to refine its meteorological utility.