A3Bench - Detailed Questions: leaderboard

Zero-shot multiple-choice video QA accuracy (%) on fine-grained detailed compositional questions in A3Bench (arXiv:2611.24762).

Metric: Detailed Questions Accuracy (%). Source: arxiv.org. Saturation forecast: Around March 2028. 14 models tracked. Updated 20 Nov 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 8B Instruct35.19#1035
2GPT-5.133.32#242
3Ovis2.5-9B33.09#1189
4Gemma 4 26B A4B (IT)30.86#561
5Qwen 3 VL 8B (Thinking)30.77#920
6InternVL3.5-8B28.75#1599
7InternVL3-9B28.22#1731
8InternVL2.5-8B26.75#2370
9Qwen 2.5 VL 7B Instruct23.73#1904
10InternVideo2.5-8B21.51
11VideoLLaMA3-7B21.41#2000
12X-CLIP20.07
13ViCLIP16.16
14LLaVA-Video-7B-Qwen213.11#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-detailed-questions · How It Works · Data refreshed daily, snapshot 2026-11-20.