HumDial-EIBench - Consistent Emotion (Chinese): leaderboard

Control set of the acoustic-semantic conflict task of HumDial-EIBench (2026): 50 single-turn Chinese human recordings whose vocal emotion and text sentiment agree, where the model identifies the emotional state of the speaker among four options. Accuracy in percent; higher is better.

Metric: Accuracy (%) of identifying the speaker's true emotional state on 50 single-turn Chinese consistent samples, where vocal emotion and text sentiment agree (four options, human-recorded); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen2.5-Omni88#1588
2Step-Audio-2-mini82
3GLM-4-Voice82
4Gemini 2.5 Flash80#463
5Kimi-Audio80#1735
6Doubao-realtime74
7GPT-4o Audio70#1129
8Freeze-Omni50

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

Interactive version: theaggregate.ai/benchmark?slug=humdial-eibench-consistent-emotion-chinese · How It Works · Data refreshed daily, snapshot 2026-11-21.