Lin Junyang, who leads the large language model team at Ali Tongyi's Qianwen division, announced that the 4B and 8B variants of the Qwen3-VL models are set for an official launch this week. These two compact visual language models are designed for ease of deployment and boast substantial application potential in mobile devices and robotics. Historically, there has been a pronounced performance disparity between smaller and larger models. However, this time around, not only have state-of-the-art large models been introduced, but also compact models that deliver performance on par with their larger counterparts. Despite having a reduced number of parameters, these smaller models exhibit remarkable spatial intelligence. They are poised to offer crucial support for the advancement of embodied intelligence and stand out as excellent substitutes for the Qwen2.5-VL models.