Multimodal Speech Understanding on MMSU Paralinguistic subset (test)
65.18AccuracyQwen2-Audio + PCLM
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2-Audio + PCLMPCLM=true2026.05 | 65.18 | |
| Qwen2-Audio + PCLM + DPOPCLM=true, Training=DPO2026.05 | 63.26 | |
| AF3 + PCLM + DPOPCLM=true, Training=DPO2026.05 | 54.78 | |
| Step-Audio-R12026.05 | 54.51 | |
| AF3 + PCLMPCLM=true2026.05 | 54.06 | |
| Gemini 2.52026.05 | 51.05 | |
| Qwen2-Audio + SFT w/o PCLMTraining=SFT, PCLM=false2026.05 | 49.41 | |
| Qwen3-Omni2026.05 | 49.13 | |
| Qwen2-Audio + SFT + DPO w/o PCLMTraining=SFT + DPO, PCLM=false2026.05 | 47.86 | |
| AF3 + SFT + DPO w/o PCLMTraining=SFT + DPO, PCLM=false2026.05 | 45.58 | |
| AF3 + SFT w/o PCLMTraining=SFT, PCLM=false2026.05 | 44.76 | |
| Kimi-Audio2026.05 | 41.48 | |
| AF32026.05 | 37.74 | |
| GPT-4o Audio2026.05 | 36.55 | |
| MiMo-Audio2026.05 | 35.64 | |
| Qwen2-Audio2026.05 | 34.37 | |
| Qwen2.5-Omni2026.05 | 33.45 | |
| VITA-Audio2026.05 | 29.54 | |
| AF22026.05 | 27.44 | |
| SALMONN2026.05 | 6.84 |