The models behind your forecast, ranked

When your local meteorologist or weather app shows next week's highs, a 10-day outlook, or a "future radar" loop a few days out, it starts with computer models like these. Forecasters talk about "the Euro," "the GFS," and "the Canadian." Those are ECMWF IFS, GFS, and GEM, and all three are on this board, next to the newer AI models. Here is how they actually score.

How the order is set: the World Meteorological Organization verification framework, anomaly correlation of 500 hPa geopotential for the extratropical Northern Hemisphere. No house model on this board. First place is stable. Everything below it rotates with the variable, the region, and the season.

Leaderboard

Global model standings

#	Model	Type	Grid

Score: 500 hPa anomaly correlation (ACC), extratropical Northern Hemisphere, deterministic runs. Skill horizon is the lead day where ACC falls to 0.80. Models with identical scores share a rank. Values are consensus figures compiled from WMO lead-centre verification and published model papers (September 2026); only entries marked pub are directly published numbers. Click a row for its full profile.

Interactive

Accuracy explorer

500 hPa anomaly correlation against lead day, days 1 to 10. Toggle models to compare. Hover or tap the chart to read every plotted model at that lead day.

Separate class

AI models

These are the newest forecasters on the board. Instead of solving the physics equations step by step, they learn from decades of past weather, and some already beat the traditional models on parts of the test. Most are not yet what your TV forecast runs on, but they are catching up fast.

Machine-learned models compete on a different footing. They share the board above, but use the Physics / AI filter to compare like with like: most AI models are trained on ERA5 and initialised from ECMWF analyses. Reported results: GenCast beat the ECMWF ensemble on 97.2% of verification targets, 99.8% past 36 hours. AIFS now posts lower error than physics models on upper-air fields. Physics models still lead on heavy precipitation scores. All three claims need their own caveats, which the sources carry.

Method

How to read this

Anomaly correlation above 0.8 counts as skillful. Below 0.6 a model has no useful skill left.
Scores move with the season and with events. A bad month, like the debated 2025 dip, does not rewrite the ranking alone.
One variable never tells the whole story. Rain, wind, and upper air each crown different winners.
Sources below. If a number here goes stale, the source is where the fresh one lives.

Receipts

Sources

Disagree with the order?

Bring a public score and we will update the board.