Envision Tianji Weather Model Tops Global and China Benchmarks

On September 9, Envision's "Tianji" physical AI weather foundation model debuted at the 2026 Inclusion·Outer Temple Conference's "Energy Storage AI Innovation Forum" in Shanghai. The model is said to provide minute-level forecasts for any region worldwide over the next 45 days, with a spatial resolution of 5 kilometers and a temporal resolution of 10 minutes.

In the international WeatherBench 2 benchmark, Tianji and Google's WeatherNext 2 were the only models to reach the top tier. Tianji led Google on key energy-related metrics such as wind speed and precipitation. Domestically, Tianji ranked first in the China National Meteorological Center's AI weather forecast model demonstration program and won first place overall in the State Grid's AI load forecasting competition.

The model leverages global meteorological data, satellite remote sensing information, and on-site measurement data from renewable energy stations, parks, and factories. Its training is supported by Envision's Mission Gobi AI computing center, and the forecasts in turn support wind and solar power generation, forming an "Energy for AI, AI for Energy" closed loop.