One model for cyclone track, intensity and size, with open weights

WeatherNext Cyclones, from Google DeepMind and Google Research, predicts a tropical cyclone's track, intensity and size together with global weather. According to the Nature paper's abstract, it was trained on global analysis data plus a global database of historical cyclones. Google AI's account contrasts this with the usual split: global physics-based models capture broad patterns, and forecasters switch to zoomed-in regional models to judge intensity.

For cyclones from 2023 to 2025, the paper reports an average lead-time advantage of a day or more over leading operational models. It generates large ensembles (many alternative forecasts for one storm) extending 15 days ahead. Google AI's post says up to 1,000 simulations per storm give forecasters a fuller range of outcomes than one most-likely path.