[key points]

Key points

• •Alibaba's Qwen-3 Omni is a multilingual open-source foundation model capable of processing text, images, audio, and video with accuracy comparable to specialized monomodal models.
• •Qwen-3 Omni's natively multimodal architecture relies on Thinker-Talker and technical improvements such as MoE, AuT, and a multi-codebook representation for diverse speech synthesis and low latency.