SSMNBench - Single-View Sufficiency: leaderboard

Metric: Mean accuracy (%) over the five single-view-sufficiency tasks (fine-grained action, human-human contact, human-object contact, relative self-pose, anatomical joint), where one view suffices; four-option multiple choice on dense, occlusion-heavy multi-view scenes of people, given exactly the annotated necessary views (the +0 setting); 300 questions per task, images resized to 1920x1080, single-letter answers. Source: arxiv.org. Saturation forecast: Around August 2028. 16 models tracked.

Top models

#ModelScore
1GPT-5.247.5
2Gemini 2.5 Pro44.6
3Gemini 2.5 Flash42.7
4Qwen 2.5 VL 72B Instruct42.7
5Qwen 2.5 VL 32B Instruct40.3
6InternVL3-78B36.2
7Qwen 2.5 VL 7B Instruct31.3

Interactive version: theaggregate.ai/benchmark?slug=ssmnbench-single-view-sufficiency · How It Works · Data refreshed daily, snapshot 2026-09-29.