Qwen3.8-LiveTranslate, the simultaneous interpretation model, has been released, with latency reduced by 18%.

Beating AI News: Qwen releases the real-time simultaneous interpretation model Qwen3.8-LiveTranslate. It reduces LAAL (Latency of Average ALignment, a measure of average lag time in simultaneous interpretation) from 2.8 seconds in the previous generation to 2.3 seconds, and adds real-time speaker diarization, synchronized source and translated text output, and long-context disambiguation.

The model can now distinguish between different speakers, map translations to specific utterances, and make translated speech more consistently preserve each speaker's voice. Source text and translated text are also returned in synchronized streaming. The model also uses prior context to determine names, terminology, and references, reducing ambiguity that arises from looking at only the current sentence.

API pricing is basically unchanged compared with the previous generation. Alibaba Cloud's price list shows that in the Beijing region, the audio input, image input, text output, and audio output prices for Qwen3.8-LiveTranslate are exactly the same as those for Qwen3.5-LiveTranslate; in the Singapore region, all four prices have slightly decreased.