SorTech-Inje University Set New Highest Record in Windows Desktop AI Agent Standard Benchmark

A joint research team comprising a domestic AI company and university achieved a top record in Microsoft's Windows Desktop AI Agent standard benchmark. SorTech's AX Frontier Team and Inje University's Media Lab completed all 154 tasks in the Windows Agent Arena (WAA) with a single clean run, achieving a 64.22% success rate (98.9/154), surpassing the previous record of OS-Symphony (GPT-5) at 63.5%. WAA, an open-source benchmark developed by Microsoft with Carnegie Mellon University and Columbia University, evaluates real-world tasks across seven domains in Windows 11 environments. The results demonstrate the effectiveness of the research team's proprietary agent operation technology, which significantly outperformed a baseline model with only safety tool integration.