OSCBench - Object State Change Accuracy: leaderboard
Metric: Object state change accuracy: whether the object reaches the target state the prompt implies (e.g. a whole apple becoming sliced), human ratings (three raters, 1-5 Likert, mean opinion score printed on a 0-1 scale, x100) of one generated video per OSCBench scenario (140 videos per model); higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Veo-3.1-Fast | 78.6 | |
| 2 | Kling 2.5 Turbo | 72.6 | |
| 3 | Wan2.2 | 56 | |
| 4 | HunyuanVideo-1.5 | 52.4 | |
| 5 | HunyuanVideo | 40.2 | |
| 6 | Open-Sora-2.0 | 38 |
Interactive version: theaggregate.ai/benchmark?slug=oscbench-object-state-change-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-11.