Tongyi Lab presents Qwen3.5-LiveTranslate for real-time interpretation

Tongyi Lab introduces Qwen3.5-LiveTranslate, a real-time interpretation model supporting 3,500 language pairs with visual anchoring and voice cloning.

Tongyi Lab, Alibaba's lab, introduces Qwen3.5-LiveTranslate, a real-time interpretation model. The tool boasts over 3,500 language pairs, very low latency, and on-the-fly speech translation, designed to keep an international audience synchronized. It supports comprehension and written output in 60 languages, with voice output available in 29 of them.

Several features set it apart. Visual anchoring integrates real-time image context to resolve ambiguities and more accurately translate what appears on screen. Voice cloning reproduces the speaker's voice in the target language, also in real time. Keyword customization finally allows for injecting specific vocabulary, such as names, brands, or industry terms, to prevent transcription errors.

Qwen3.5-LiveTranslate is accessible via Qwen Omni, with an API rollout planned for later.