ArchesWeatherGen: a generative AI model to tackle meteorological uncertainty

Presented in an article published in Science Advances in April 2026, ArchesWeatherGen is a new weather forecasting model based on generative artificial intelligence. It can make forecasts on a worldwide scale and requires 20 to 50 times less computational power for training than its competitors. This breakthrough project was developed by ARCHES, a joint project team at the Sorbonne University Inria Centre (Versailles-Saint-Quentin-en-Yvelines University, Sorbonne University, CNRS, Inria).