VocalCoachBench - Triplet Ranking: leaderboard
Triplet ranking in VocalCoachBench (Bang et al., 2026), a benchmark of expert vocal-coaching feedback for singing built from 515 recordings annotated by 18 professional vocal trainers. For three performances of the same song the audio-language model compares each pair, and pairwise accuracy in percent measures agreement with the ranking the trainers induce, where random choice scores 50; higher is better.
Metric: Pairwise accuracy (%; for three performances of the same song the model compares each pair, A vs B, A vs C and B vs C, and each preference is scored against the ranking induced by professional vocal trainers; random choice scores 50; main structured-prompt protocol, deterministic decoding where supported). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Omni Plus | 74.5 | #237 |
| 2 | Qwen3.5 Omni Flash | 72.4 | #548 |
| 3 | Gemini 3 Flash (Preview) | 71.2 | #141 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vocalcoachbench-triplet-ranking · How It Works · Data refreshed daily, snapshot 2026-11-12.