KOpenAudioBench - KWebQ: leaderboard
Spoken Korean short-answer questions transferred from WebQuestions, part of KOpenAudioBench (Kim et al., 2026). Accuracy in percent over 970 TTS-synthesized items; higher is better.
Metric: Accuracy (%; short-answer questions transferred from WebQuestions; 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 | 45.3 | |
| 2 | HyperCLOVA X 8B Omni | 38.5 | |
| 3 | Fun-Audio-Chat | 35.4 | |
| 4 | Qwen2.5-Omni | 35.1 | #1499 |
| 5 | MiniCPM-o 4.5 | 30 | |
| 6 | Step-Audio-2-mini | 27.3 | |
| 7 | InteractiveOmni | 5.9 | |
| 8 | Audio Flamingo 3 | 5.3 |
Interactive version: theaggregate.ai/benchmark?slug=kopenaudiobench-kwebq · How It Works · Data refreshed daily, snapshot 2026-11-12.