KVoiceBench - KOpenBookQA: leaderboard
Spoken Korean multiple-choice science questions transferred from OpenBookQA, part of KVoiceBench (Kim et al., 2026). Accuracy in percent over 445 TTS-synthesized items; higher is better.
Metric: Accuracy (%; multiple-choice science questions transferred from OpenBookQA; 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 | 74.8 | |
| 2 | Qwen2.5-Omni | 31 | #1499 |
| 3 | Fun-Audio-Chat | 28.3 | |
| 4 | HyperCLOVA X 8B Omni | 27.4 | |
| 5 | MiniCPM-o 4.5 | 25.2 | |
| 6 | Step-Audio-2-mini | 7.4 | |
| 7 | InteractiveOmni | 7.4 | |
| 8 | Audio Flamingo 3 | 6.5 |
Interactive version: theaggregate.ai/benchmark?slug=kvoicebench-kopenbookqa · How It Works · Data refreshed daily, snapshot 2026-11-12.