KVoiceBench - KMMSU: leaderboard
Spoken Korean multiple-choice questions transferred from MMSU, part of KVoiceBench (Kim et al., 2026). Accuracy in percent over 3,006 TTS-synthesized items; higher is better.
Metric: Accuracy (%; multiple-choice questions transferred from MMSU; Korean spoken questions synthesized with TTS; single run at temperature 0.7). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Raon-Speech | 55.5 | |
| 2 | Fun-Audio-Chat | 28.7 | |
| 3 | Qwen2.5-Omni | 28.3 | #1499 |
| 4 | MiniCPM-o 4.5 | 26.9 | |
| 5 | HyperCLOVA X 8B Omni | 24 | |
| 6 | Step-Audio-2-mini | 11.7 | |
| 7 | Audio Flamingo 3 | 10 | |
| 8 | InteractiveOmni | 7 |
Interactive version: theaggregate.ai/benchmark?slug=kvoicebench-kmmsu · How It Works · Data refreshed daily, snapshot 2026-11-12.