Pangu-Weather

Published in Nature in July 2023 ('Accurate medium-range global weather forecasting with 3D neural networks'), Pangu-Weather was Huawei Cloud's breakthrough AI forecasting system and the first AI model to beat the operational ECMWF IFS on deterministic medium-range forecasts. Its 3D Earth-Specific Transformer treats height as a true third dimension to capture relationships across pressure levels, and a hierarchical temporal aggregation strategy cuts cumulative forecast error; a 24-hour global forecast runs in about 1.4 seconds on a single V100 — over 10,000 times faster than the IFS.

Huawei has not released Pangu-Weather's weights publicly, so it remains a research milestone rather than a usable model. It also kicked off the modern AI-weather race that GenCast, GraphCast, and NeuralGCM continued, and its deterministic single-forecast design was later complemented by ensemble approaches that better quantify forecast uncertainty.