WBench - Subject Action: leaderboard
The subject action score of WBench (2026), a multi-turn benchmark for interactive video world models: text-driven video world models extend a scene turn by turn under their own controls. The score is subject-action interaction on the non-navigation split, on a 0-100 scale; higher is better.
Metric: Score (0-100): subject-action interaction on the non-navigation split, text-driven models only, multi-turn generation; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kling 3.0 | 85.6 | |
| 2 | Wan2.7 | 83.4 | |
| 3 | Seedance 1.5 | 80 | |
| 4 | HY-Video 1.5 | 55.6 | |
| 5 | LTX-2.3 | 51.8 | |
| 6 | LongCat-Video | 48.4 | |
| 7 | YUME 1.5 | 47 | |
| 8 | Cosmos 2.5 | 41.6 | |
| 9 | Kairos 3.0 | 41.4 |
Interactive version: theaggregate.ai/benchmark?slug=wbench-subject-action · How It Works · Data refreshed daily, snapshot 2026-11-22.