Google Open-Sources WeatherNext AI for Cyclone Forecasting

Google DeepMind published its latest research in Nature and officially open-sourced the code and model weights for its AI weather model, WeatherNext. The model achieves state-of-the-art (SOTA) accuracy in predicting the track, intensity, size, and wind field structure of cyclones (typhoons/hurricanes), capable of forecasting cyclone formation and dynamics up to 15 days in advance. Data shows it provides an average of 24 hours more lead time compared to previous best models—a leap equivalent to roughly a decade of progress in traditional meteorology—holding immense value for disaster prevention, mitigation, and saving lives.

Confirmed

Why it matters

2026-08-06 ~ 2026-08-07 · 9 related posts

Primary sources

2 near-duplicate retellings: gaganghotra_ · aigclink