Envision’s Tianji Weather Model Wins ECMWF AI Weather Quest, Sets New Global Standard for Physics‑AI Forecasting

Overview

Envision, the Shanghai‑based green‑technology group, announced that its home‑grown weather foundation model Tianji placed first overall in the European Centre for Medium‑Range Weather Forecasts (ECMWF) AI Weather Quest. The competition, which pits roughly 20 leading research teams against each other, evaluates continuous, real‑time forecasts for a 19‑to‑25‑day horizon across several key variables.

Competition Details

• Metrics evaluated: variable‑averaged near‑surface temperature, mean sea‑level pressure, and precipitation.
• Forecast window: 19‑25 days (subseasonal range).
• Evaluation method: weekly submissions are compared with observed weather; the leaderboard is updated publicly in real time.
• Participants: about 20 top teams from academia and industry worldwide.

Envision’s press release states that Tianji topped the leaderboard on *all* assessed metrics, outperforming the nearest rivals by a measurable margin. ECMWF has not released the exact score differentials, but the public leaderboard confirms Tianji’s lead.

Technical Highlights

• Physics‑AI hybrid: Tianji is described as the world’s first model that tightly couples physical atmospheric equations with large‑scale AI training.
• Data scale: the model was trained on petabyte‑scale datasets that combine satellite, radar, aircraft, and ground‑based observations – a “space‑air‑ground” three‑dimensional network.
• Computing power: high‑performance computing clusters enable rapid generation of forecasts from minute‑level nowcasting to subseasonal trends.
• Performance gains: reported improvements in wind‑speed and solar‑irradiance accuracy for key renewable‑energy regions, though exact error‑reduction percentages were not disclosed.

Real‑World Implications

• Energy sector: more reliable wind and solar forecasts can help grid operators schedule generation, reduce reliance on fossil‑fuel backup, and lower operating costs.
• Weather‑sensitive industries: agriculture, aviation, and logistics can benefit from better week‑ahead temperature and precipitation outlooks.
• Policy and research: the result validates the physics‑AI approach, encouraging further investment in hybrid models for climate‑scale prediction.

How to Verify the Claim

1. ECMWF AI Weather Quest leaderboard – accessible on the ECMWF website, shows weekly rankings and metric scores.
2. Envision’s official press release – published on PR Newswire (Oct 30 2026) and mirrored on Envision’s corporate newsroom.
3. Independent benchmarks – WeatherBench 2 and China’s National Meteorological Center demo program results are publicly available for cross‑checking.

Outlook

Subseasonal forecasting (two weeks to two months ahead) remains one of the most uncertain segments of meteorology. Envision’s claim that Tianji reduces that uncertainty is promising, but independent peer‑reviewed studies will be needed to quantify the improvement across diverse climate regimes. If the model’s performance holds up, it could become a cornerstone for next‑generation renewable‑energy planning and climate‑risk assessments.

--- *All statements are based on Envision’s announcement and publicly available ECMWF competition data. No proprietary figures were disclosed.*