Report
OSWorld benchmark spans 26 custom iOS apps for personally intelligent phone agents
A post about the COLM 2026 presentation says the apps share a persistent user identity and interconnected personal data.
TLDR
A post describes OSWorld as an interactive benchmark for personally intelligent phone agents, built around a persistent user identity across 26 custom iOS apps. It says the apps include analogs of OpenTable, Uber, DoorDash, Airbnb and Chase, with interconnected messages, transactions, travel histories, social relationships and preferences.
Combined views
3.3K
1 Source, first seen ago