Definition
Plain language
A harder benchmark of real phone tasks used to test mobile AI agents.
As stated in the literature
A mobile-agent benchmark of multi-step, often cross-application phone tasks, used alongside AndroidWorld; terminal-based agents substantially outperform screen-based ones on it.
Why it matters: Its tough multi-app tasks reveal which agent designs actually hold up on realistic phone work rather than toy single-screen actions.
For example, it might ask an agent to find a contact in one app and use that information to book something in another.
Heard on the show
“7 — hits that seventy-one point eight percent on the benchmark called AndroidWorld, and just under fifty-two percent on a harder one called MobileWorld.”Episode 157 — When an AI Coding Agent Drives a Phone Through the Terminal, No Screen Needed