Google DeepMind WeatherNext: open cyclone AI with a day of extra warning
In a Nature paper, Google DeepMind and Google Research showed WeatherNext achieves state-of-the-art accuracy on cyclone track, intensity, and wind structure—on average gaining more than a full day of lead time (three-day forecasts matching prior two-day skill), roughly a decade of meteorological progress. The team is open-sourcing WeatherNext Cyclones (used in the 2025 hurricane season, including NHC support on Hurricane Melissa), WeatherNext 2 (later operationalized update), and WeatherNext 2-mini (single-TPU Colab). Models use Functional Generative Networks for up to 1,000-member ensembles from ~28 km inputs, with forecasts exploreable on Weather Lab as part of Google Earth AI.






