Qwen ships real-time speech interpreter cutting lag to 2.3s

Qwen released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model built on what Qwen calls an Interleave architecture. The company reports it reduces average lagging, its measure of translation delay after someone finishes speaking, from 2.8 seconds to 2.3 seconds across 60 languages, alongside unquantified improvements to faithfulness, fluency, and conciseness. New capabilities include real-time speaker diarization, distinguishing speakers in multi-party conversations while preserving each speaker's voice through voice cloning, a synchronized bilingual display showing source and translation together, and a form of long-context disambiguation that uses conversation history to keep names and terminology consistent across a session.