General Conversation on VoiceBench
4.19AlpacaEval ScoreStep-Audio-2
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Step-Audio-2Model variant=Base2026.04 | 4.19 | 3.12 | 3.36 | 55.15 | 52.8 | 50.82 | 68.13 | 58.53 | 39.64 | 92.88 | 64.15 | |
| Step-Audio-2Deep thinking training=w/ think, Training data ratio=1:12026.04 | 4.08 | 4.03 | 3.79 | 51.9 | 44.48 | 51.61 | 65.49 | 56.31 | 17.4 | 95.58 | 63.62 | |
| Step-Audio-2Deep thinking training=w/ think, Training data ratio=1:0.52026.04 | 3.98 | 3.94 | 3.69 | 49.73 | 44.85 | 53.04 | 71.87 | 54.69 | 18.83 | 100 | 64.21 | |
| Step-Audio-2Deep thinking training=w/o think, Training data ratio=1:12026.04 | 3.77 | 3.75 | 3.42 | 48.28 | 39.24 | 47.69 | 68.79 | 50.25 | 23.61 | 84.62 | 59.72 | |
| Step-Audio-2Deep thinking training=w/o think, Training data ratio=1:0.52026.04 | 3.38 | 3.43 | 3.02 | 49.73 | 38.34 | 36.88 | 56.7 | 50.66 | 20.74 | 87.69 | 54.8 |