Google DeepMind's WeatherNext AI Model Open Sourced After Cyclone Forecasting Breakthrough

Google DeepMind and Google Research have published a paper in Nature detailing their WeatherNext AI model, which achieves state-of-the-art accuracy in predicting tropical cyclone track, intensity, and wind structure. The model provides forecasters with an extra day of lead time, equivalent to a decade of meteorological progress. During the 2025 hurricane season, WeatherNext assisted the National Hurricane Center (NHC) in forecasting Hurricane Melissa's rapid intensification and landfall in Jamaica, enabling advance warnings. The model is now open sourced, along with WeatherNext 2 and WeatherNext 2-mini, to empower the research community. WeatherNext uses Functional Generative Networks (FGNs) to generate ensembles of 1,000 predictions, capturing rare events like rapid intensification. It operates at a coarse resolution of 28x28km, 100x coarser than traditional models, yet outperforms them. The model was co-trained on global atmospheric data and the IBTrACS database of nearly 5,000 historical storms. Collaborators include the NHC, Cooperative Institute for Research in the Atmosphere (CIRA), and the UK Met Office. The open source release includes code, model weights, and access via Weather Lab, part of Google Earth AI.