HumDial-EIBench - Causal Reasoning: leaderboard

Implicit causal reasoning on HumDial-EIBench (2026): from real human-recorded multi-turn dialogues from the ICASSP 2026 HumDial Challenge, an audio language model infers the unstated cause of the current emotion of the user, among four options whose distractors are synthesized in the voice of the speaker. Accuracy in percent, mean of the Chinese and English sets; higher is better.

Metric: Accuracy (%), mean of the Chinese and English accuracies, on Task 2 implicit causal reasoning (infer the unstated cause of the user's current emotion) four-option multiple-choice questions built from real human-recorded dialogues from the ICASSP 2026 HumDial Challenge test set, audio language models answer from the audio, with three adversarial distractors synthesized in the speaker's voice; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Flash79.67#463
2Qwen2.5-Omni76.99#1588
3GPT-4o Audio76.47#1129
4Step-Audio-2-mini71.84
5Kimi-Audio68.71#1735
6GLM-4-Voice55.66
7Doubao-realtime52.97
8Freeze-Omni24.02

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-causal-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.