EmoTrans - Emotion State Identification: leaderboard

EmoTrans (2026) emotion state identification: the model names each target person's emotion before and after a transition in a real-world video clip; accuracy (%) over 724 questions, higher is better.

Metric: Accuracy (%) on Emotion State Identification (724 questions: each person's emotion before and after the change), on EmoTrans, built from 1,000 manually annotated real-world video clips (OpenHumanVid and CH-SIMS v2) over 12 scenarios with one or more target persons; zero-shot, video with its text (and audio where the model accepts it); higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Seed 2.0 Lite41.2#300
2GPT-5.440.5#135
3Qwen 3.5 122B A10B38.8#346
4Qwen 3.5 27B38.4#375
5Gemini 3 FlashNon-reasoning38.3#175
6GLM-4.5V38#754
7GLM-4.6V37.4#710
8Gemini 3 Pro36.5#145
9Qwen 2.5 VL 72B Instruct33.1#841
10Qwen 3 VL 32B Instruct31.2#606
11InternVL3.5-14B30.8#1504
12Gemini 2.5 FlashNon-reasoning26.8#463
13Qwen 3 VL 8B24.4#917
14Qwen 3 Omni 30B A3B23.8#847
15InternVL3.5-8B21.8#1604

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=emotrans-emotion-state-identification · How It Works · Data refreshed daily, snapshot 2026-11-21.