- Published on
Alibaba Qwen Team Releases Qwen3.8-LiveTranslate: A Real-Time Interpretation Model That Cuts Average Lag to 2.3 Secon...
- Authors
- Name
- DeepHorus
Alibaba's Qwen team has released Qwen3.8-LiveTranslate, a next-generation real-time simultaneous interpretation model that listens to live speech and returns translated text and audio while the speaker is still talking. The model is available as a hosted API on Alibaba Cloud Model Studio and QwenCloud under the model ID qwen3.8-livetranslate-flash-realtime. It also accepts optional video frames to improve accuracy in noisy environments.
Simultaneous interpretation has always involved a tradeoff between waiting for more context and reducing listener delay. Qwen rebuilt this loop with a new Interleave architecture, which the team says improves faithfulness, fluency, and conciseness. The key latency metric, Length-Adaptive Average Lagging, dropped from 2.8 seconds to 2.3 seconds, an approximately 18% reduction in average lag.
The release adds three major capabilities. Real-time speaker diarization distinguishes speakers in multi-party conversations and preserves each voice through stable voice cloning. Synchronized bilingual display streams source text alongside the translation, while long-context disambiguation uses conversation history to keep names and terminology consistent throughout a session.
The model understands 60 languages and can speak 29 of them, returning both audio and text; the remaining 31 languages return text only. Inputs include audio and optional images, with visual cues such as lip movements and gestures helping in noisy rooms. The documentation recommends sending no more than two images per second and configuring up to 1,000 hotwords for fixed term mappings.
Developers connect through the WebSocket Realtime API, with default audio at 16 kHz PCM input and 24 kHz PCM output. Singapore list pricing per one million tokens is $7.50 for audio input, $0.55 for image input, $20 for text output, and $30 for audio output; Beijing pricing is lower. One hour of speech input and output costs about $1.54 in Singapore before text and image tokens.
The context window is 53,248 tokens, with 49,152 for input and 4,096 for output, and default rate limits are 10 requests and 100,000 tokens per minute. Built on the Qwen-Omni stack, the Flash version also supports offline audio and video translation. This launch could make real-time interpretation more practical for meetings, live events, and multilingual customer support.
