A3Bench - Overall: leaderboard
Zero-shot multiple-choice video QA benchmark (arXiv:2611.24762) measuring question-count-weighted accuracy (%) on aggressive action recognition and aggressor/victim/bystander role-action binding.
Metric: Overall Accuracy (%). Source: arxiv.org. Saturation forecast: Around June 2028. 14 models tracked. Updated 20 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 46.96 | #1035 |
| 2 | InternVL2.5-8B | 45.77 | #2370 |
| 3 | Ovis2.5-9B | 43.16 | #1189 |
| 4 | Gemma 4 26B A4B (IT) | 42.72 | #561 |
| 5 | InternVL3-9B | 42.55 | #1731 |
| 6 | GPT-5.1 | 41.3 | #242 |
| 7 | InternVL3.5-8B | 41.06 | #1599 |
| 8 | InternVideo2.5-8B | 40.7 | |
| 9 | Qwen 3 VL 8B (Thinking) | 39.37 | #920 |
| 10 | VideoLLaMA3-7B | 38.22 | #2000 |
| 11 | Qwen 2.5 VL 7B Instruct | 37.3 | #1904 |
| 12 | ViCLIP | 23.82 | |
| 13 | X-CLIP | 20.12 | |
| 14 | LLaVA-Video-7B-Qwen2 | 18.5 | #1977 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=a3bench-overall · How It Works · Data refreshed daily, snapshot 2026-11-20.