Ali QianWen has officially released its state-of-the-art simultaneous interpretation model, Qwen3.8-LiveTranslate. Leveraging the innovative Interleave architecture, this model achieves real-time interpretation reconstruction with an impressive average character delay of just 2.3 seconds, while supporting an extensive range of 60 languages. In addition to these capabilities, the model introduces three cutting-edge features: real-time speaker separation for multi-speaker scenarios, seamless in-frame display of both original text and its translation, and advanced long-context disambiguation to ensure accuracy in complex conversations.

The Qwen3.8-LiveTranslate model adopts a sophisticated Hybrid MoE Thinker–Talker dual-module design, which enables it to outperform industry-leading real-time simultaneous interpretation systems and previous-generation models across multiple performance metrics. This superiority is evident in evaluations conducted on the Omnilingua-MSpeaker multi-speaker long audio dataset and the FLEURS audio test set, where the model demonstrated exceptional accuracy, fluency, and adaptability.