Alibaba releases Qwen3.8-LiveTranslate: simultaneous interpretation model with 2.3s latency
千问大模型 · wechat · 2026-09-19
Alibaba's Qwen team officially released Qwen3.8-LiveTranslate, a simultaneous interpretation model that uses an Interleave architecture to restructure interpretation as a single interleaved audio-text stream, cutting average latency (LAAL) from 2.8s to 2.3s while supporting 60 languages.
Built on a HybridMoE Thinker-Talker design, it adds real-time speaker diarization with stable voice cloning, frame-aligned bilingual output, and long-context disambiguation for names and terminology. On the 14-direction Omnilingua-MSpeaker benchmark it beats mainstream real-time interpretation systems on fidelity, fluency, conciseness and diarization error rate; on FLEURS it leads across 70 language directions in quality, latency, ASR and TTS. Future plans include pushing latency limits, cross-session memory and more long-tail languages. Demo and API are live.
More from Models
- Open-weight science model Hydrogen-1.0 beats GPT-6 Astra on FSR at 72% lower cost — AKAMA199 · 2026-11-04
- Community asks Moonshot when Kimi K3.1 model weights will drop — bgmshana · 2026-11-04
- Meta's Muse Spark Open license counts distilled models as derivatives — aidenclarke_12 · 2026-11-04
- Tencent Hy4 trails GPT-6.1 on hard tests as Kimi K3.1 Pro beats it on two benchmarks — heyshrutimishra · 2026-11-04
- Tencent's Hy4 agent self-trained a smaller model that beat Tencent's own baseline — heyshrutimishra · 2026-11-04
- Hy4 tops open models in finance eval at 3.04, hallucination rate cut to 5.45% — heyshrutimishra · 2026-11-04