Qwen launches real-time interpretation model for 60 languages
Qwen has released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model that processes live speech and optional video frames. It produces translated text and speech while the speaker continues talking. The model uses a new Interleave architecture, which Qwen says improves faithfulness, fluency and conciseness. Reported length-adaptive average lagging fell from 2.8 seconds to 2.3 seconds.
The release adds real-time speaker diarization for multi-party conversations, more stable voice cloning, synchronized source and translated text, and long-context disambiguation for names and terminology. Visual input can use lip movements, gestures and on-screen text to help with ambiguous speech or noisy environments. Users can also configure hotwords with fixed target translations.
Qwen3.8-LiveTranslate understands 60 languages and provides audio as well as text for 29 of them; the other 31 produce text only. It is available through the WebSocket Realtime API on Alibaba Cloud Model Studio and QwenCloud under the model ID qwen3.8-livetranslate-flash-realtime. Singapore list pricing includes $7.50 per 1M audio-input tokens and $30 per 1M audio-output tokens. The context window is 53,248 tokens.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.