Google DeepMind WeatherNext cyclone forecasting models

Google DeepMind's WeatherNext AI model gives cyclone forecasters an average of 24 additional hours of lead time compared to existing methods, meaning three-day forecasts are as accurate as prior models' two-day forecasts. The research was published in Nature on August 6, 2026. Google DeepMind described that gain as corresponding to 'roughly a decade's worth of meteorological progress'. During the 2025 hurricane season, WeatherNext predicted with 80% confidence that Hurricane Melissa would make landfall in Jamaica as a Category 5 storm five days before it did. The model was co-trained on nearly 20 terabytes of global atmospheric data and the IBTrACS historical cyclone database covering nearly 5,000 storms. It uses Functional Generative Networks to produce ensemble forecasts, scaling to 1,000 predictions per run, to cover rare events such as rapid intensification. WeatherNext Cyclones operates at a 28x28km resolution, which Google DeepMind said is 100 times coarser than traditional models require. Google DeepMind open-sourced WeatherNext 2, WeatherNext Cyclones, and WeatherNext 2-mini on GitHub under commercially usable licenses. WeatherNext 2-mini runs on a single TPU in a free public Colab notebook.