Google DeepMind has developed a new AI model called GraphCast that demonstrates high accuracy in global weather forecasting. The model, described in a paper published in Science, can predict weather conditions up to 10 days in advance more accurately and much faster than the current gold-standard system, the High Resolution Forecast (HRES) from the European Centre for Medium-Range Weather Forecasts (ECMWF). GraphCast is a machine learning model trained on decades of historical weather data. It uses a graph neural network architecture to process data across the globe’s surface, making predictions at a resolution of approximately 25 kilometers. In tests, it outperformed HRES on over 90% of 1,380 metrics. The model’s speed is a significant advantage, generating a 10-day forecast in under a minute on a single Google TPU v4 machine, compared to hours of computation on a supercomputer for traditional physics-based models. Researchers emphasize that AI models like GraphCast are not replacements but powerful complementary tools for existing numerical weather prediction systems. The code for GraphCast has been made publicly available. For the full details, read the complete article at https://www.science.org/doi/10.1126/science.adl2336.