Which Weather Model Is Most Accurate? We Score Them Every Day, in the Open

VIDRAFT has released Darwin-Prediction-Weather and a public Hugging Face leaderboard that evaluates major AI and physics-based weather models daily across 24 cities, four weather variables, and forecast horizons of up to 10 days. Using ERA5 reanalysis as the reference and root-mean-square error as its metric, the leaderboard compares models from organizations including ECMWF, NOAA, DWD, JMA, ECCC, and MET Norway under identical conditions. Its 2026 results through November 7 report that ECMWF’s AIFS and NOAA’s AIGFS outperform their respective physics-based counterparts, while VIDRAFT’s skill-weighted fusion system, Darwin-Prediction-Weather, records the lowest errors for maximum and minimum temperature, wind speed, and precipitation from one to seven days ahead. The system combines global forecasts with local corrections and a decision-oriented model, and it also reports leading results on the WeatherBench 2 2020 deterministic benchmark. Forecast records, observed values, model files, and the leaderboard are publicly available, while planned additions include longer forecast horizons, station-based verification, more locations and variables, and a public forecasting competition.

No tracked trend matches for this post yet.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.