Google DeepMind and Google Research have released their latest generation of AI weather forecasting models, and the performance gap between machine learning and traditional supercomputer-driven meteorology just got a lot wider.
The updated WeatherNext family, including WeatherNext 2 and a specialized cyclone-tracking variant called WeatherNext Cyclones (WN-C), delivers forecasts up to 8 times faster than its predecessors while outperforming conventional physics-based systems on nearly every measurable dimension. WeatherNext 2 beats previous models on 99.9% of forecast variables.
What the models actually do
WeatherNext operates at 0.25-degree spatial resolution, which translates to roughly 30 kilometers at the equator. Forecasts initialize every six hours, at 00, 06, 12, and 18 UTC, and ensemble versions extend predictions out to 15 days.
The cyclone-specific model, WN-C, is where things get especially interesting. A peer-reviewed paper published in Nature on August 6, 2026, documents WN-C’s performance during the 2025 Atlantic hurricane season. The headline finding: the AI model provides approximately one additional day of lead time over conventional forecasting systems for cyclone track, intensity, and size predictions. The model proved particularly effective at predicting Hurricane Melissa, one of the more notable storms of the 2025 season.
