Google DeepMind’s WeatherNext AI model extends tropical cyclone warning lead times by 24 hours and is now freely available as an open source tool for global researchers.
On 6 August 2026, Google DeepMind published a research in Nature demonstrating state-of-the-art accuracy in predicting tropical cyclone track, intensity, and wind structure using its WeatherNext AI model. The system provides an extra day of warning, delivering three-day cyclone forecasts that match the accuracy previously achieved at two days.
WeatherNext connects wind patterns all over the world with smaller storm actions in one single model. It was trained using 20 terabytes of weather data and the IBTrACS database of past storms. The system uses Functional Generative Networks (FGNs) to create 1,000 possible weather predictions. This helps it catch dangerous weather events like sudden storm growth. It can process a 15-day weather forecast in less, than a minute using a Tensor Processing Unit (TPU).
Unlike traditional physics-based systems requiring fine-grained inputs, WeatherNext Cyclones operates efficiently on a coarser 28 x 28 kilometre input resolution (111 x 111 kilometres for its mini variant). Tested operationally during the 2025 Atlantic hurricane season, the National Hurricane Center (NHC) used the model to forecast Hurricane Melissa’s rapid intensification and landfall in Jamaica.
Google DeepMind open sourced the code and model weights via GitHub under an Apache 2.0 licence across three variants: WeatherNext Cyclones, WeatherNext 2, and WeatherNext 2-mini (capable of running in a free Google Colab notebook). Live outputs are accessible on the refreshed Weather Lab interactive platform within Google Earth AI.
The research was co-developed by Google DeepMind and Google Research in collaboration with the National Hurricane Center, the Cooperative Institute for Research in the Atmosphere (CIRA), and the UK Met Office.
