DeepMind's WeatherNext model achieves breakthrough forecasting cyclones
Points and comments are a snapshot, not live.
DeepMind's WeatherNext model offers an extra day's lead time for cyclone forecasts, now open-sourced.
Published in Nature on August 6, 2026, WeatherNext achieves state-of-the-art accuracy in predicting cyclone track, intensity, and wind structure. Its three-day forecasts match prior models' two-day accuracy, equivalent to a decade of meteorological progress. The model uses Functional Generative Networks to run 1,000-member ensembles in under a minute on a TPU, capturing tail risks. It was co-trained on global atmospheric data and the IBTrACS database. During the 2025 hurricane season, it helped the NHC predict Hurricane Melissa's rapid intensification. DeepMind is open-sourcing WeatherNext 2 and WeatherNext Cyclones on GitHub.
What commenters are saying
Commenters appreciated the practical value of improved cyclone forecasting for saving lives and resources, particularly for maritime scheduling and evacuation planning. Some discussed the difficulty of predicting earthquakes due to sparse data on long seismic cycles. Others noted the model's potential for cargo ships to save fuel and improve safety. A minor thread joked about using the technology for faster transatlantic flights.