Qwen3.5-Omni adds real-time voice, video tools

This reads like Qwen continuing to compress the gap between “chat model” and “full assistant stack.” The important part is not just that it handles more modalities, but that it packages those modalities with tools and low-latency speech, which is where a lot of demos stop short.

• –Native multimodal support across text, image, audio, and video makes it suitable for richer assistant workflows.
• –Real-time voice interaction and voice cloning are the most product-defining features here, not just table stakes extras.
• –Web search and function calling make it more useful for agentic workflows than a pure perception model.
• –Long-context audio/video understanding suggests it is aiming at meetings, media analysis, and interactive copilots.
• –The main question is execution quality: omni models often look broad on paper, but latency, reliability, and cross-modal consistency decide whether they are actually usable.