MOSS-Live 六项全双工语音指标领先，多轮对话仍落后 GPT-Live-1
MOSS-Live Leads Six Full-Duplex Voice Metrics but Trails GPT-Live-1 on Multi-Turn Tasks
想给应用加语音交互的开发者值得看，MOSS-Live 抢话指标最强、背景噪音答错率只有 2%，但多轮对话比不过 GPT-Live-1，音频理解输给 Doubao，选型前能省不少试错成本。
复旦 MOSS 团队旗下 OpenMOSS 发布全双工语音模型 MOSS-Live，在六项全双工语音评测指标上排名第一，其中在背景人声干扰下答错的错误率仅为 2%，抗噪能力是其核心优势。
评测同时揭示了短板：在多轮任务（multi-turn tasks）上，MOSS-Live 不及 OpenAI 的 GPT-Live-1；在音频理解维度则落后于字节跳动 Seed 团队的 Doubao-Seed-Duplex。这意味着 MOSS-Live 在抢话、插话等全双工交互指标上占优，但在需要持续理解上下文的复杂对话场景仍有差距。
全双工语音是当前对话模型竞争的关键方向，GPT-Live-1、Doubao-Seed-Duplex 与 MOSS-Live 分别代表美中两国的不同技术路线，分项指标的分化也为开发者按场景选型提供了直接参照。
MOSS-Live Leads Six Full-Duplex Voice Metrics but Trails GPT-Live-1 on Multi-Turn Tasks
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.
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include MOSS-Live, GPT-Live-1, Doubao, speech models.
Primary source verification: published directly by pandaily with first-party authority.
