AffectSim (Reference View): leaderboard

Metric: Macro-F1 (%; P-Ref: an informative camera trajectory chosen by a privileged reference observer from body pose and scene geometry; five-class emotion recognition (angry, fearful, happy, neutral, sad) on the 3,941 test episodes of AffectSim, full-body affective motions replayed in Habitat indoor scenes; frozen recognizer, eight uniformly sampled frames for most open models, full video or eight frames for closed models). Source: arxiv.org. Saturation forecast: Around 2029. 24 models tracked.

Top models

#ModelScore
1Claude Opus 535.68
2GPT-5.6 Sol31.48
3Gemini 3 Pro30.21
4InternVL2.5-78B30.02
5Gemini 3 Flash28.33
6GPT-5.228.2
7Qwen 3 VL 8B Instruct15.54
8Qwen2.5-Omni-7B12.09
9Qwen 3 VL 8B (Thinking)11.15
10MiniCPM-V-2.69.89
11Qwen 2.5 VL 72B Instruct7.57
12Qwen 2.5 VL 7B Instruct6.12

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