MOSS-Live debut: full-duplex voice AI beats GPT Realtime 2.1 on interruption benchmarks
机器之心 · wechat · 2026-11-02
MOSS Intelligence (模思智能) launched MOSS-Live, a full-duplex real-time interaction system that can listen while speaking, handle interruptions via semantic turn detection, recover context, call tools asynchronously mid-conversation, and support streaming simultaneous interpretation. Online demo and limited-free API are open.
Vendor benchmarks claim wins over GPT Realtime 2.1, GPT-Live-1 and Doubao Seed Duplex: 1.44% false interruption rate (vs 3.83% for GPT Realtime 2.1), only 2% false responses to irrelevant speech (others: 68–88%), 98.17% backchannel robustness, and 51% tool-call success. Architecture pairs a foreground interaction model with a background reasoning/task-execution model; demos cover meetings, in-car assistance, video calls and cross-language chat.
The company also recently raised a 1B+ RMB Pre-A round, open-sourced MOSS-Image (3M+ downloads), and has MOSS-TTS 2.0 in preview.
Related event: Fudan-linked MOSS team releases MOSS-Live full-duplex voice AI(2 posts)→
More from Models
- Claude's charming support menu: vent, think-it-thru or rest mode — repligate · 2026-11-03
- Miles Brundage: Claude Code has gotten pushy with suggested follow-up projects — Miles_Brundage · 2026-11-03
- Hand-assigning every weight: blogger builds a working transformer with zero training — voooooogel · 2026-11-03
- ComfyUI User Details How Flux2Klein and Qwen 2.1 Fail at Character Height Differences — Ishelle91 · 2026-11-03
- Community rebuilds edited Qwen3.5-0.8B from logs, scores 51.0% on MATH-500 — _akhaliq · 2026-11-03
- Claude's alignment answers aren't literal defaults but a posture of supplying the right answer — 1a3orn · 2026-11-03