AI looks at blurry 28km pixels, somehow predicts hurricanes better than physics models that needed 100x the resolution

Google DeepMind's WeatherNext AI model achieves state-of-the-art accuracy in predicting tropical cyclone track, intensity, and wind structure, delivering roughly a decade's worth of forecasting improvement: three-day forecasts now match what prior models could only do for two days. The model was co-trained on 20 TB of global atmospheric data and nearly 5,000 historical storms, runs 1,000-member ensembles in under a minute on a TPU, and already helped the NHC issue a historic warning for Hurricane Melissa in 2025. Google is open sourcing WeatherNext 2, WeatherNext Cyclones, and a compact WeatherNext 2-mini that runs in a free Colab notebook.