Massive Multimodal Audio Reasoning on MMAR
56.4Normalized Judge ScoreStep-Audio-R1
Evaluation Results
| Method | Links | |
|---|---|---|
| Step-Audio-R12025.08 | 56.4 | |
| AF3Full name=Audio Flamingo 32025.08 | 50.5 | |
| VIRBABackbone=Qwen2.5-Omni2025.08 | 49.8 | |
| Kimi-Audio2025.08 | 45.2 | |
| VIRBABackbone=Qwen2-Audio2025.08 | 42.7 | |
| Qwen2.5-Omni2025.08 | 38.9 | |
| Single-view RL2025.08 | 36 | |
| GLM-4-Voice2025.08 | 34.8 | |
| Qwen2-Audio base2025.08 | 31.4 |