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
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 | 35.68 |
| 2 | GPT-5.6 Sol | 31.48 |
| 3 | Gemini 3 Pro | 30.21 |
| 4 | InternVL2.5-78B | 30.02 |
| 5 | Gemini 3 Flash | 28.33 |
| 6 | GPT-5.2 | 28.2 |
| 7 | Qwen 3 VL 8B Instruct | 15.54 |
| 8 | Qwen2.5-Omni-7B | 12.09 |
| 9 | Qwen 3 VL 8B (Thinking) | 11.15 |
| 10 | MiniCPM-V-2.6 | 9.89 |
| 11 | Qwen 2.5 VL 72B Instruct | 7.57 |
| 12 | Qwen 2.5 VL 7B Instruct | 6.12 |
Interactive version: theaggregate.ai/benchmark?slug=affectsim-reference-view · How It Works · Data refreshed daily, snapshot 2026-09-29.