DeepMind Open-Sources WeatherNext: Over 24 Hours of Extra Cyclone Predictive Accuracy

gaganghotra_ · x · 2026-08-07

Google DeepMind announced the open-sourcing of WeatherNext, releasing its code and model weights on GitHub for researchers and meteorological agencies.

The model achieves state-of-the-art accuracy in predicting cyclone track, intensity, and wind structure, giving forecasters more than 24 hours of additional predictive accuracy. Furthermore, it can generate 1,000 possible cyclone scenarios in under a minute on a single TPU, helping meteorologists assess everything from likely paths to rare, dangerous rapid intensification events.

Related event: Google Open-Sources WeatherNext AI for Cyclone Forecasting(9 posts)→

Original post →

More from Models

Models channel →