FineBench (Subset) - Object Manipulation: leaderboard
The object manipulation questions of FineBench (Faure et al., 2026) on the paper's representative subset of 7 movie clips (20,143 questions in all), the split on which proprietary models were run: four-option questions about how a designated person handles objects. Accuracy in percent (chance 25%); higher is better.
Metric: Accuracy (%; four-option multiple-choice questions on how a designated person handles objects; representative subset of 7 AVA movie clips, 20,143 questions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 86.3 | #702 |
| 2 | GPT-5 Mini (2025-08-07) | 85.3 | #341 |
| 3 | MiniCPM-V-2.6 | 84.8 | #3095 |
| 4 | Gemini 1.5 Flash | 81.9 | #1671 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=finebench-subset-object-manipulation · How It Works · Data refreshed daily, snapshot 2026-11-12.