Google DeepMind open-sourced its WeatherNext AI model to predict cyclone tracks, intensity, and wind structure. The journal Nature published the model's state-of-the-art accuracy results.
WeatherNext provides an extra day of predictive accuracy on average. This improvement represents roughly a decade of meteorological progress. The National Hurricane Center utilized the model for real-world operations in 2025.
The model maintains high performance while using lower-resolution data. Google released the code and model weights to accelerate global weather innovation.
A compact version called WeatherNext 2-mini runs on a single TPU in a free public notebook. This tool helps researchers predict severe weather and mitigate economic losses.