Alibaba's Qwen3.5-LiveTranslate translates live with voice cloning and 3,500+ pairs

Alibaba's Tongyi Lab unveiled Qwen3.5-LiveTranslate, a next-generation real-time interpretation model designed for seamless global communication. It supports an impressive 3,500+ language pairs with ultra-low latency, enabling near-instant translation as users speak. A standout feature is real-time voice cloning, which replicates the speaker's unique voice across 29 languages, preserving tone and emotion. The model also integrates visual context analysis to eliminate ambiguity from on-screen content, and allows dynamic injection of custom vocabulary (names, brands, domain terms) to prevent transcription drift. Qwen3.5-LiveTranslate handles comprehension and writing across 60 languages, making it a comprehensive solution for live events, customer support, and multilingual content creation.

Alongside the translation model, Tongyi Lab released Qwen3.7-Max, its latest flagship designed for the agent era. The model excels in coding (frontend prototypes, multi-file refactors, real debugging), long-horizon autonomy (35-hour kernel optimization with 1,000+ tool calls), and scaffold-agnostic deployment (works with Claude Code, OpenClaw, Qwen Code, or custom stacks). Qwen3.7-Max scored 56.6 on the Artificial Analysis Intelligence Index (up 4.8 points from its preview), with gains in scientific reasoning, agentic capability, and coding. Preview versions also landed on the Arena leaderboards, ranking Alibaba #5 for vision and #6 for text among labs.