Speech Understanding on AIR-Bench
29.9SERSALMONN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SALMONNSize=7B2026.03 | 29.9 | 34.3 | 42.3 | |
| GPT-4o AudioSize=-2026.03 | 51.2 | 61.6 | 89.3 | |
| Audio-ThinkerSize=8.4B2026.03 | 56.2 | 67.5 | — | |
| Gemini 2.5 FlashSize=-2026.03 | 56.4 | 68.5 | 88.6 | |
| Qwen2.5-omni + Reasoning SFTSize=7B2026.03 | 58.5 | 68.1 | 85 | |
| Audio Flamingo 3Size=7B2026.03 | 59.5 | 76.8 | 79.6 | |
| Qwen2.5-omni + Only STASize=7B2026.03 | 59.5 | 66 | 84.3 | |
| Qwen2.5-omniSize=7B2026.03 | 60.2 | 63.9 | 83.5 | |
| Audio-ReasonerSize=8.4B2026.03 | 60.5 | 56.3 | 88.1 | |
| TGSRSize=7B2026.03 | 62.5 | 70.4 | 89.3 |