FineBench - Object Manipulation: leaderboard
The object manipulation questions of FineBench (Faure et al., 2026): four-option questions generated from AVA's human-annotated action labels about how a designated person handles objects in densely annotated movie clips. Accuracy in percent over the full set of 64 clips (chance 25%); higher is better.
Metric: Accuracy (%; four-option multiple-choice questions on how a designated person handles objects; all 64 AVA movie clips, 199,420 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 73.9 | #1842 |
| 2 | MiniCPM-V-2.6 | 72.8 | #3095 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=finebench-object-manipulation · How It Works · Data refreshed daily, snapshot 2026-11-12.