Alibaba Tongyi Lab has open-sourced its GUI-Owl-1.5 and Mobile-Agent-v3.5 model families, designed to autonomously interact with desktop, mobile, and browser interfaces. Built on the Qwen3-VL foundation, these models come in six sizes ranging from 2B to 32B parameters, optimized for low latency or advanced reasoning.

They achieve state-of-the-art results on over 20 GUI-agent benchmarks, including OSWorld-Verified and AndroidWorld. Key innovations include a hybrid data flywheel for trajectory generation, unified Chain-of-Thought synthesis for integrated reasoning, and multi-platform reinforcement learning for robust performance.

These models represent a significant step toward fully autonomous AI agents capable of human-like interface manipulation.

Source: https://github.com/X-PLUG/MobileAgent and modelscope.cn/models/iic/GUI-Owl-1.5-8B-Instruct