An AI weather model from Envision has ranked first overall in the latest AI Weather Quest, a forecasting competition run by the European Centre for Medium-Range Weather Forecasts (ECMWF). The Shanghai-based green technology company announced on 30 October that its Tianji weather foundation model, competing under the codename Arcana, beat nearly 20 teams worldwide and took first place across all assessed metrics for the 19–25 day forecast window.
Envision says the codename Arcana means “mysteries” or “knowledge.” The company describes Tianji as the “world’s first physics-AI weather model,” a claim that comes from Envision itself rather than from ECMWF.
How the AI Weather Quest tests an AI weather model
The AI Weather Quest is not a one-off benchmark. It evaluates models through continuous, real-time forecasting and verification, so each prediction is recorded before the weather it describes has happened. Teams submit forecasts every week, and organizers compare them with the weather that follows, publicly updating a leaderboard.
The competition looks beyond the usual 15-day medium-range horizon. It focuses on probabilistic forecasts for days 19 to 25 of three things:
- Weekly mean near-surface temperature
- Weekly mean sea level pressure
- Weekly accumulated precipitation
Temperature and precipitation are scored over grid cells that are mainly land, while mean sea level pressure is assessed globally. According to Envision, Tianji led on the variable-averaged score as well as on near-surface temperature, mean sea level pressure and precipitation.
Why subseasonal forecasting is hard
Forecasts several weeks ahead sit in a difficult gap between day-to-day weather prediction and seasonal climate outlooks. “Subseasonal forecasting — two weeks to two months out — is one of the hardest challenges in meteorology,” said Robert Zheng, head of Envision’s Tianji Weather Foundation Model. “The longer the horizon, the greater the uncertainty in atmospheric changes, and the higher the demand on a model’s ability to identify evolving weather trends.”
Zheng said the model builds on Envision’s experience in new energy and draws on petabyte-scale data with high resolution in space and time, along with a “space–air–ground” three-dimensional data network.
Earlier results for the Tianji AI weather model
The ECMWF result follows a series of other rankings that Envision has reported for Tianji:
- A top-tier placement in the global WeatherBench 2 evaluation, which Envision says put it on par with Google’s WeatherNext2 model
- First place overall in the deterministic forecast track of the AI Weather Forecast Model Demonstration Program run by China’s National Meteorological Center
- First place in State Grid’s AI load forecasting competition
- First place in a business competition run by China Southern Power Grid
Envision says the latest win shows Tianji’s capability across the full chain, from short-term forecasting to subseasonal prediction.
What it means for energy and planning
For a green technology company working in new energy, better weather prediction has direct commercial value. Envision says Tianji, powered by high-performance computing, delivers fast short-term and global forecasts with significantly improved accuracy for wind speed and solar irradiance in key regions. It covers time scales from minute-level nowcasting to subseasonal trend assessment, supporting both real-time monitoring and long-term planning.
As a writer, I find the shift from short-range forecasting to weeks-ahead prediction the most interesting part of this race, because that is where planning decisions for power grids and renewables are made. For more on long-range prediction, see our stories on El Niño forecasts a year ahead, India’s new climate services forum and NYU’s AI wildfire spread prediction.
FAQ
What is Envision Tianji?
Tianji is a weather foundation model developed by Envision. The company describes it as a physics-AI weather model that covers forecasts from minute-level nowcasting to subseasonal trends.
What did Tianji win?
Competing as Arcana, it ranked first overall in ECMWF’s latest AI Weather Quest, ahead of nearly 20 teams, and led every assessed metric for the 19–25 day window.
How is the AI Weather Quest judged?
Teams submit probabilistic forecasts weekly, and ECMWF scores them against the weather that actually follows, updating a public leaderboard.