Baichuan-Audio-Instruct: benchmark results
Provider: Baichuan. Access: Open.
Aggregate Rating 1484 ± 50, rank #2613 of 10916 rated models, from 13 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SonicBench - Duration | 52 | Accuracy (%; sound-event duration (TAU); 100 recognition and | 62.9 |
| SonicBench - Reverberation | 53 | Accuracy (%; reverberation (SRIR room impulse responses); 10 | 62.9 |
| RAIL - Acquired Knowledge | 58.28 | LLM-as-judge score (%) on acquired-knowledge tasks over spee | 46 |
| SonicBench - Texture | 68 | Accuracy (%; sound texture (ESC-50); 100 recognition and 100 | 43.8 |
| SonicBench | 51 | Accuracy (%; mean over 12 physical attributes of 100 recogni | 38.7 |
| SonicBench - Brightness | 51 | Accuracy (%; spectral brightness (NSynth); 100 recognition a | 38.7 |
| SonicBench - Timbre | 52 | Accuracy (%; instrument timbre (NSynth); 100 recognition and | 38.7 |
| SonicBench - Counting | 49 | Accuracy (%; number of sound events (PicoAudio); 100 recogni | 33.9 |
| SonicBench - Loudness | 51 | Accuracy (%; loudness (integrated LUFS contrasts of at least | 33.9 |
| RAIL - Fluid Reasoning | 43.79 | LLM-as-judge score (%) on fluid-reasoning tasks (induction, | 32 |
| RAIL - Memory | 40.34 | LLM-as-judge score (%) on auditory memory tasks (memory span | 32 |
| SonicBench - Pitch | 49 | Accuracy (%; pitch height (NSynth notes, one-semitone contra | 16.1 |
Interactive version: theaggregate.ai/model?slug=baichuan-audio-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.