FineBench - Person Interaction: leaderboard
Metric: Accuracy (%; four-option multiple-choice questions on a designated person's interactions with other people; all 64 AVA movie clips, 199,420 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 63.8 |
| 2 | MiniCPM-V-2.6 | 56.5 |
Interactive version: theaggregate.ai/benchmark?slug=finebench-person-interaction · How It Works · Data refreshed daily, snapshot 2026-09-26.