GraphCast

GraphCast, published by Google DeepMind in Science on November 14, 2023, was the first AI weather model to beat conventional forecasting in a peer-reviewed head-to-head: it outperformed ECMWF's HRES on 90% of 1,380 tested metrics including wind speed, temperature, and air pressure. The graph neural network runs on a multi-mesh icosahedral representation of the globe, producing precise 10-day forecasts in under a minute on a single machine — versus hours on a supercomputer for physics-based systems. It famously predicted Hurricane Lee's Nova Scotia landfall nine days ahead.

Limitations: deterministic rather than probabilistic, so it doesn't quantify forecast uncertainty the way GenCast's ensembles do. DeepMind positioned it as a complement to traditional methods rather than a replacement, and operational use still leans on physics-based systems for initialization data.