Google DeepMind's WeatherNext Cyclones won the 2026 Gizmodo Science Fair award for advanced hurricane forecasting and is recognized for state-of-the-art accuracy in predicting tropical cyclone track, intensity, and wind structure, as supported by a Nature study and ongoing assessments.

The project currently collaborates with the National Hurricane Center, CIRA at Colorado State University, and the U.K. Met Office, with plans to expand to global partnerships to broaden impact.

WeatherNext Cyclones is open-source and designed to run on minimal technology, enabling use by communities with limited resources to improve storm safety.

The improvement is described as roughly equivalent to about a decade of progress in traditional forecasting methods.

Public real-world validation remains strongest in the Atlantic and North Pacific regions, though evidence supports performance across other basins.

Rising global temperatures could boost tropical cyclone intensity and rainfall, underscoring the importance of improved forecasts in a warming climate.

WeatherNext Cyclones should be used as an added layer of guidance alongside existing forecasts, not as a replacement.

The model can generate up to 1,000 possible scenarios to better capture low-probability, high-impact events, addressing the gray swan problem in weather extremes.

Key researchers include Ferran Alet and Tom Andersson from DeepMind, among a broad team, with ongoing efforts to scale and refine the model.

AI cannot remove inherent weather uncertainty, but it can push forecast accuracy forward and become a powerful tool in the forecasting toolbox.

WeatherNext Cyclones leverages extensive storm data to detect rare extreme events, which are increasing due to climate change.

The model is open-sourced and accessible without a supercomputer, with GitHub hosting and Weather Lab providing access to the latest versions to democratize advanced forecasting for vulnerable regions.