88Z AI, 20 September — Alibaba Tongyi has released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model supporting 60 languages with reduced latency of 2.3 seconds per character. Utilizing a Hybrid MoE architecture, it features real-time speaker diarization and long-context disambiguation, outperforming mainstream systems in multilingual evaluations.
At a glance
- Latency reduced to 2.3 seconds per character with support for 60 languages and real-time speaker diarization
- Uses Hybrid MoE dual-module architecture, leading mainstream systems in multiple performance metrics
The update
Alibaba Tongyi has officially launched Qwen3.8-LiveTranslate, a large model designed for real-time simultaneous interpretation. The model achieves significant performance improvements, reducing average latency per character from 2.8 seconds to 2.3 seconds while supporting translation across 60 languages. New features include real-time speaker diarization, synchronized output of source and target text within the same frame, and enhanced disambiguation capabilities for long contexts to ensure translation accuracy. Technically, Qwen3.8-LiveTranslate utilizes a Hybrid MoE architecture with dual Thinker–Talker modules. In evaluations involving multi-speaker long audio and the FLEURS multilingual benchmark, the model demonstrates superior performance in key metrics such as faithfulness, fluency, and latency. It outperforms current mainstream simultaneous interpretation systems, providing more efficient technical support for real-time cross-language communication.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 千问大模型Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.