AudioBench - CoVoST2 EN-ZH BLEU: leaderboard
AudioBench CoVoST2 EN-ZH BLEU speech-translation BLEU slice measuring translation quality between the named source and target languages.
Metric: Mean Score (%). Source: github.com. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Phi-4 Multimodal Instruct | 45.3 | #906 |
| 2 | SeaLLMs-Audio-7B | 36.45 | #1018 |
| 3 | SALMONN-7B | 33.89 | #1293 |
| 4 | Qwen2-Audio-7B-Instruct | 25.77 | #880 |
| 5 | Whisper Large-v3 | 16.41 | #1271 |
| 6 | Qwen-Audio-Chat | 15.33 | #1244 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=audiobench-covost2-en-zh-bleu · How It Works · Data refreshed daily, snapshot 2026-10-20.