WeatherNext Cyclones

WeatherNext Cyclones is a machine-learning model built for one job: predicting where a tropical cyclone will go, how strong it will get and what its wind field will look like. Google DeepMind released it on 6 August 2026 alongside two general-purpose forecasting models, WeatherNext 2 and WeatherNext 2-mini. The result DeepMind leads with is lead time. Three-day forecasts from the model are about as accurate as two-day forecasts were before it — roughly 100 km of track error at day three, ahead of the ECMWF ensemble — and intensity error comes in around 11 knots, better than the specialised HWRF hurricane model. DeepMind frames the extra day as comparable to about a decade of conventional progress, which in a field that improves in small increments is an unusual claim to make. The method is what makes it interesting. The model runs on a grid of roughly 28 by 28 kilometres, about a hundred times coarser than the physics-based models it is measured against; a hurricane eye is smaller than a single cell. It is not simulating the storm at all. It has learned, from history, what storms of a given shape and trajectory tend to do next. DeepMind says the work was done with the US National Hurricane Center, the Cooperative Institute for Research in the Atmosphere and the UK Met Office, and points to the 2025 season — Hurricane Melissa's rapid intensification and Jamaica landfall — as operational use. Code and model weights are published on GitHub, forecasts are visualised in Weather Lab, and the models feed Google's Earth AI tooling.

▸News

DeepMind opens its cyclone forecaster: three-day tracks as good as yesterday's two-day ones

8/6/2026 · Google DeepMind

▸Videos

No videos yet.

▸Reviews