Alibaba's MAI-UI Confirmed as Leading Mobile Agent Framework
Researchers from Alibaba's Tongyi Lab have established a new state-of-the-art in mobile automation with their MAI-UI framework, which has achieved top scores on the rigorous MobileWorld benchmark.
MobileWorld is designed to test GUI agents in realistic, complex scenarios involving multi-application workflows and long-horizon tasks. MAI-UI achieved an impressive 41.7% success rate on the benchmark, significantly outperforming the closest competitors and previous baselines.
"MAI-UI demonstrates a 20.8 point lead over the strongest competitor, showcasing the robust capability of Tongyi's vision-language models in understanding and navigating mobile interfaces."
The framework leverages the Qwen3-VL model family and excels in tasks ranging from standard GUI interactions to complex agent-user dialogues. The release confirms the rapid maturation of mobile agent technology, paving the way for more autonomous and helpful smartphone assistants.