Envision's Tianji Weather Foundation Model Tops ECMWF's AI Weather Quest, Setting a New Global Benchmark for Physics-AI Forecasting

Envision's Tianji weather foundation model has reportedly achieved the top position in the latest AI Weather Quest, an AI weather competition organized by the European Centre for Medium-Range Weather Forecasts (ECMWF). This competition specifically evaluates probabilistic forecasts for weekly mean temperature, mean sea level pressure, and accumulated precipitation for a 19–25 day forecast window, a period considered part of the challenging subseasonal forecasting domain.

The reported success against nearly 20 international teams across all assessed metrics indicates a notable capability in predicting weather trends several weeks in advance. Robert Zheng, head of Envision's Tianji Weather Foundation Model, characterized subseasonal forecasting, two weeks to two months out, as one of the hardest meteorological challenges due to the increasing uncertainty with longer horizons. The AI Weather Quest's continuous, real-time evaluation against actual weather conditions suggests a robust assessment of sustained performance.

Tianji is described by Envision as the world's first physics-AI weather model, leveraging petabyte-scale high spatiotemporal resolution data and a "space–air–ground" three-dimensional data network. Beyond this competition, Tianji has reportedly demonstrated its value in energy applications, including winning AI load forecasting competitions for State Grid and China Southern Power Grid, underscoring its practical utility for real-time monitoring and long-term planning across various time scales.