Unveiling ‘MAI-UI’: The Revolutionary AI for Superior Smartphone Automation, Outperforming Gemini 3 Pro

Alibaba’s AI research team, Tongyi-MAI, has introduced MAI-UI, an advanced AI capable of automatically operating smartphones, outperforming the Gemini 3 Pro in benchmark tests. MAI-UI is categorized into four models: MAI-UI-2B, MAI-UI-8B, MAI-UI-32B, and MAI-UI-235B-A22B. Notably, the MAI-UI-32B model achieved an impressive score of 73.5% on the ScreenSpot-Pro benchmark, surpassing Gemini 3 Pro’s performance. MAI-UI excels at complex tasks, such as booking train tickets and managing calendar events, and supports multifunctional capabilities through cloud and local execution. The lightweight MAI-UI-2B can run locally on smartphones, while the higher-performance MAI-UI-32B operates through the cloud. Additionally, the automatic model selection feature optimizes performance by choosing the best option for user instructions. Models MAI-UI-2B and MAI-UI-8B are accessible on Hugging Face under the Apache license 2.0. For further details, check the technical report.

Source link