KoALa-Bench - KCSAT: leaderboard
KoALa-Bench (2026) evaluates large audio language models on Korean speech understanding and on faithfulness to the speech context, under clean and noisy audio. This board is accuracy in percent on the long-form listening questions of the Korean College Scholastic Ability Test, clean audio; higher is better.
Metric: Accuracy (%) on the long-form listening questions of the Korean College Scholastic Ability Test (2006-2012 recordings, multiple choice), Korean speech synthesized or recorded per item, clean audio (first value of each clean / noisy pair), mean over four prompts with greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash Lite | 81.18 | #964 |
| 2 | Qwen3 Omni 30B A3B Instruct | 80.59 | #930 |
| 3 | Voxtral-Mini-3B-2507 | 69.41 | #3682 |
| 4 | GPT-Audio Mini | 52.9 | #918 |
| 5 | Raon-Speech-9B | 45.88 | |
| 6 | Gemma 3n E4B (IT) | 33.53 | #2401 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=koala-bench-kcsat · How It Works · Data refreshed daily, snapshot 2026-11-21.