EmoS: leaderboard
Seven-way emotion recognition from short video clips (EmoS, Guo et al., 2026): English MELD and Chinese CH-SIMS v2 clips re-annotated by three annotators per clip and aggregated with Dawid-Skene, scored on the 800-clip high-confidence test subset. Accuracy in percent for multimodal LLMs prompted zero-shot, with answers outside the seven labels counted wrong; higher is better.
Metric: Accuracy (%; Basic-7 emotion label (anger, disgust, fear, joy, neutral, sadness, surprise) of a video clip from the high-confidence test subset, human labels aggregated by Dawid-Skene; zero-shot multimodal LLM, answers outside the label set count as wrong; English and Chinese clips pooled). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 61.1 | #135 |
| 2 | Qwen2.5-Omni-7B | 57.1 | #1797 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=emos · How It Works · Data refreshed daily, snapshot 2026-11-12.