AffectSim (Reference View): leaderboard
AffectSim (2026) tests emotion recognition from body movement in simulated homes: emotion-labelled full-body motions are replayed in Habitat indoor scenes and a frozen vision-language model classifies each of 3,941 test episodes as angry, fearful, happy, neutral or sad. The observation is an informative camera trajectory chosen by a privileged reference observer from body pose and scene geometry. Score is macro-F1 in percent; higher is better.
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 July 2028. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 35.68 | #22 |
| 2 | GPT-5.6 Sol | 31.48 | #30 |
| 3 | Gemini 3 Pro | 30.21 | #137 |
| 4 | InternVL2.5-78B | 30.02 | #1014 |
| 5 | Gemini 3 Flash | 28.33 | #164 |
| 6 | GPT-5.2 | 28.2 | #176 |
| 7 | Qwen 3 VL 8B Instruct | 15.54 | #981 |
| 8 | Qwen2.5-Omni-7B | 12.09 | #1789 |
| 9 | Qwen 3 VL 8B (Thinking) | 11.15 | #887 |
| 10 | MiniCPM-V-2.6 | 9.89 | #3063 |
| 11 | Qwen 2.5 VL 72B Instruct | 7.57 | #846 |
| 12 | Qwen 2.5 VL 7B Instruct | 6.12 | #1833 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=affectsim-reference-view · How It Works · Data refreshed daily, snapshot 2026-11-13.