AnyAudio-Judge Bench - Chinese (Holistic): leaderboard
AnyAudio-Judge Bench (2026) tests audio-language models as judges: given a speech, sound, music or mixed-audio clip and a text instruction, the model decides whether the audio follows it, on balanced aligned and misaligned pairs (chance 50). This board is accuracy in percent on the Chinese instruction set with holistic prompting (one decision per pair), averaged over seven subsets; higher is better.
Metric: Classification accuracy (%) on the Chinese instruction subset under holistic prompting (one match or mismatch decision per pair): judge a speech, sound, music or mixed-audio clip against a text instruction as aligned or misaligned, balanced 1:1 positives and negatives (chance 50); accuracy averaged over the seven subsets (Speech-Real, Speech-Gen, Sound-Real, Sound-Gen, Music-Real, Music-Gen, Mix); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 80.01 | #266 |
| 2 | Qwen3 Omni 30B A3B Instruct | 60.24 | #904 |
| 3 | Qwen2.5-Omni-7B | 51.76 | #1789 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=anyaudio-judge-bench-chinese-holistic · How It Works · Data refreshed daily, snapshot 2026-11-13.