Beyond the GUI Paradigm: Do Mobile Agents Need the Phone Screen?

The article presents a comparative evaluation of mobile agents, specifically focusing on command-line interfaces (CLI) versus graphical user interfaces (GUI). Three coding agents—Claude Code, Terminus-2, and mini-swe-agent—were assessed on two benchmark datasets, AndroidWorld and MobileWorld, with Claude Code achieving the highest performance at 71.8% and 51.9%, respectively, surpassing all GUI baselines. The introduction of the CLI-Advantage Task Suite, which includes 45 task templates, highlights the efficiency of CLI agents in handling complex user intents, suggesting significant potential for further advancements in mobile CLI applications.