multi-task audio reasoning on MMAU mini
0.742Sound ScoreAudioRouter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AudioRouterBackbone=Qwen2.5-Omni Based2026.02 | 0.742 | 0.769 | 0.823 | 0.778 | |
| Omni-CLSTBackbone=Qwen2.5-Omni Based2026.02 | 0.685 | 0.724 | 0.808 | 0.739 | |
| AudioRouterBackbone=Qwen2-Audio Based2026.02 | 0.638 | 0.672 | 0.712 | 0.674 | |
| Qwen2.5-OmniBackbone=Qwen2.5-Omni Based2026.02 | 0.624 | 0.681 | 0.78 | 0.695 | |
| R1-AQABackbone=Qwen2-Audio Based2026.02 | 0.601 | 0.654 | 0.737 | 0.664 | |
| Qwen2-AudioBackbone=Qwen2-Audio Based2026.02 | 0.512 | 0.586 | 0.69 | 0.596 | |
| Random Guess2026.02 | 0.267 | 0.246 | 0.267 | 0.26 |