Participate in the evaluation for the first time and reach the top directly! Vivo Blue Heart Xiao V ranks first in the mobile GUI smartphone list

Kuai Technology reported on August 21 that SuperCLUE announced the AgentCLUE Mobile mobile GUI Agent benchmark evaluation list. This is an evaluation specifically designed to test the working ability of mobile phone AI agents. Vivo Blue Heart Xiao V participated in the evaluation for the first time and directly won the first place in the overall list with a total score of 92.11 points, widening the score gap. There are a total of 6 products participating in this review, all of which are mainstream mobile phone GUI smartphones on the market. The second place is Open AutoGLM of Zhipu AI, with a score of 76.75 points. The third place came from Alibaba Mobile Agent 3.5, scoring 73.45 points. The fourth place is also Alibaba’s MAI UI, the fifth place is GELab Zero, and ByteDance UI TARS is ranked sixth. The test score is divided into three core dimensions, namely complex intent disassembly and dynamic planning, multi-modal GUI perception and understanding, and cross-application full-link execution. Blue Heart Xiao V ranked first in all three categories, scoring 94.41 points for complex intent disassembly, 91.69 points for multi-modal GUI perception, and 89.12 points for cross-application full-link execution. Simply put, whether it is understanding complex instructions, understanding the mobile phone interface, or completing tasks continuously across multiple apps, its performance is at the leading level. In terms of calling methods, Lanxin Xiao V, Open AutoGLM, and Mobile Agent 3.5 use API calls, and the remaining three products use local deployment mode. API calls can gain greater capabilities from large cloud models. The advantage of local deployment is that it does not require an Internet connection and the data remains on the local machine. Both options have their own trade-offs. GUI Agent is an AI assistant on your mobile phone. You just tell us your needs, and it automatically recognizes the screen interface, clicks buttons, switches software, and completes a complete set of operations across APPs without us having to click manually step by step. For example, let it help you search for movies, order takeout, organize your schedule, and let the agent automatically run the process. Of course, the mobile GUI agent is not perfect yet. Execution failures will still occur when encountering interface changes or special pop-ups. It can also be seen from this evaluation that each company’s