Envision, a global leader in green technology, has achieved a major milestone in meteorological artificial intelligence. The company's independently developed weather foundation model, Envision Tianji, has secured the first-place ranking overall in the latest edition of the AI Weather Quest. Organized by the European Centre for Medium-Range Weather Forecasts (ECMWF), this highly competitive event evaluated nearly 20 top teams from across the globe. Competing under the codename Arcana—signifying “mysteries” or “knowledge”—the Tianji model outperformed its global peers across all assessed metrics, establishing a new benchmark for subseasonal weather prediction.

The Mechanics of the AI Weather Quest

The AI Weather Quest is designed to test the limits of predictive models through continuous, real-time forecasting and rigorous verification. Rather than relying on static datasets, the competition assesses sustained performance under constantly changing, real-world weather conditions.

While standard medium-range forecasts typically focus on a 15-day horizon, this competition specifically targets the subseasonal window, evaluating probabilistic forecasts for days 19–25. The evaluation focuses on several critical metrics:

  • Variable-averaged, near-surface temperature
  • Mean sea level pressure
  • Weekly accumulated precipitation

To ensure a comprehensive assessment, temperature and precipitation are evaluated over grid cells that are predominantly land-covered, while mean sea level pressure is assessed on a global scale. This dual approach tests a model's ability to anticipate weekly conditions and manage atmospheric uncertainty several weeks in advance. Participating teams submit their forecasts on a weekly basis, and the organizers evaluate these submissions against subsequent actual weather observations to update the public leaderboard.

Technical Innovation: Physics-AI Integration

Predicting weather patterns weeks in advance has long been considered one of the most formidable challenges in meteorology. As the forecasting horizon extends, the chaotic nature of the atmosphere introduces compounding uncertainties, making traditional numerical models less reliable.

To address this challenge, Envision developed what it describes as the world's first physics-AI weather model. According to Robert Zheng, head of Envision's Tianji Weather Foundation Model, subseasonal forecasting—spanning two weeks to two months out—demands an exceptional ability to identify evolving weather trends amidst high atmospheric uncertainty.

The architecture of Tianji relies on two primary pillars:

  • Petabyte-Scale High Spatiotemporal Resolution Data: The model is trained on massive datasets that capture detailed atmospheric variables over time and space.
  • Space–air–ground Three-Dimensional Data Network: This integrated network feeds real-time, multi-layered observational data into the model, bridging the gap between physical meteorological principles and deep learning capabilities.

By combining physical laws with advanced artificial intelligence, Tianji can deliver reliable forecasts across full time scales, from short-term outlooks to subseasonal predictions.

Broader Industry Implications and Track Record

The victory at the ECMWF AI Weather Quest is the latest in a series of high-profile achievements for the Tianji model. The foundation model has consistently demonstrated its capabilities across both academic benchmarks and practical industrial applications.

Prior to this competition, Tianji achieved a top-tier placement in the global WeatherBench 2 evaluation, performing on par with the Google WeatherNext2 Model. Additionally, the model secured first place overall in the deterministic forecast track of the AI Weather Forecast Model Demonstration Program, which was organized by China's National Meteorological Center.

Beyond theoretical meteorology, Envision has successfully applied Tianji to the energy sector, where precise weather forecasting is critical for grid stability and renewable energy integration. The model has won first place in two major power grid competitions:

  • The State Grid AI load forecasting competition
  • The China Southern Power Grid business competition

These achievements highlight the practical utility of Tianji's full-chain forecasting capabilities. In an era where renewable energy sources like wind and solar are highly dependent on weather conditions, the ability to accurately predict atmospheric changes weeks in advance is crucial for managing grid loads and optimizing green energy distribution.