MOSS-Live 在六项全双工指标中排名第一,但有 2% 的时间会错误地回应背景语音。在多轮任务上,它落后于 GPT-Live-1,在音频理解上落后于豆包。MOSS-Live tops six full-duplex metrics, wrongly answering background speech 2% of the time. It trails GPT-Live-1 on multi-turn tasks and Doubao on audio understanding.