PAWBench (World Modeling) - Calibration: leaderboard

Metric: Conditional total-variation distance between the produced and reference outcome distributions for video generators, times 100, on a 0 to 100 scale (0-100); lower is better; PAWBench measures whether a system reproduces the reference distribution over physically possible outcomes across 50 rollouts on 25 scenarios per physical track, averaged over scenes that pass the readability gate. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.

Top models

#ModelScore
1Cosmos 3 Super I2V20.5
2MiniMax H324.2
3Wan2.726.3
4Wan2.226.3
5LTX-2.330.1
6LTX-2.530.2
7Seedance 230.5
8Kling 3 Std.34.9
9Veo3.1 Fast35.4
10LingBot-Video-MoE41.8

Interactive version: theaggregate.ai/benchmark?slug=pawbench-world-modeling-calibration · How It Works · Data refreshed daily, snapshot 2026-09-26.