PitchBench - Pitch under Time Stretching: leaderboard
Pitch hearing in audio-language models (PitchBench, Liessens Dujardin et al., 2026): naming the sounding pitch of a tone after resampling or pitch-preserving time stretching (192 stimuli), with synthetic stimuli. A pitch answer counts when any of MIDI number, scientific pitch notation or frequency is right; accuracy in percent, higher is better.
Metric: Accuracy (%; name the sounding pitch after resampling or time stretching). Source: arxiv.org. Saturation forecast: Around June 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Omni Plus | 88.5 | #237 |
| 2 | Qwen3.5 Omni Flash | 79.2 | #548 |
| 3 | Gemini 3.1 Pro (Preview) | 18.2 | #91 |
| 4 | Gemini 3 Flash | 9.4 | #159 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pitchbench-pitch-under-time-stretching · How It Works · Data refreshed daily, snapshot 2026-11-12.