iOSWorld: A Benchmark for Personally Intelligent Phone Agents
2026-06-08 • Machine Learning
Machine LearningComputation and Language
AI summaryⓘ
The authors created iOSWorld, a new test platform that simulates an iPhone with a fake user who has a history, preferences, and personal data across 26 apps. This helps measure how well AI agents can handle tasks involving multiple apps and remember personal information, unlike older tests that only use simple, isolated instructions. They tested various AI models and found that even the best could only complete around half the tasks, struggling especially with multi-app challenges. The authors also found that giving agents access to extra information helps bigger models but not smaller ones. They made iOSWorld available for others to use and improve AI on real phone-like experiences.
iOS simulatorAI agentpersonalizationmulti-app tasksuser identityprivileged visionXML accessibility treebenchmarkinteractive taskscomputer-use models
Authors
Lawrence Keunho Jang, Mareks Woodside, Geronimo Carom, Andrew Keunwoo Jang, Jing Yu Koh, Ruslan Salakhutdinov
Abstract
A useful phone agent needs to be personally intelligent. It should reason over a user's identity, history, and preferences as they exist on the device, not just follow isolated instructions in an impersonal sandbox. Existing mobile agent benchmarks lack this kind of personalization. We introduce iOSWorld, the first interactive native iOS simulator benchmark built around a persistent user identity spanning 26 newly built iOS apps. These apps contain connected data such as transactions, messages, travel records, social relationships, and financial activity. iOSWorld includes 133 tasks across three increasingly difficult categories. Single-app tasks (27) test one app, multi-app tasks (60) span 2 to 8 apps, and memory and personalization tasks (46) require agents to infer patterns from personal data. We evaluate frontier and open-source computer-use models in both vision-only and privileged vision+XML settings. The best configuration reaches 52\% overall but only 37\% on multi-app tasks. Privileged vision+XML access improves frontier models by up to 26 percentage points, while smaller models do not benefit from added accessibility-tree input. We release iOSWorld as an open-source benchmark with all apps, seeded data, tasks, rubrics, and evaluation code.