E3mo-Bench (Evoked) - Open-Vocabulary Recognition (VAD Similarity): leaderboard
E3mo-Bench (E³mo-Bench, 2026) evoked-emotion subset, about the emotion a video evokes in the viewer: the model names the emotions freely, scored with the redundancy-aware set similarity in valence-arousal-dominance space, which penalizes distant confusions more, in percent; higher is better.
Metric: Redundancy-aware VAD set similarity F-VAD (%; predicted and reference open-vocabulary emotion sets compared in a z-normalized valence-arousal-dominance lexicon space after merging near-duplicate emotions, so that distant confusions cost more; 16 MLLMs on videos from eight emotion datasets, each assigned one affective perspective; subset: the viewer's evoked emotion). Source: arxiv.org. Saturation forecast: Around 2030. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 55.38 | #123 |
| 2 | InternVL3-8B | 50.54 | #1688 |
| 3 | Qwen2.5-Omni-7B | 49.36 | #1789 |
| 4 | Claude Sonnet 4.6 | 48.76 | #161 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=e3mo-bench-evoked-open-vocabulary-recognition-vad-similarity · How It Works · Data refreshed daily, snapshot 2026-11-13.