Benchmark RadarAI BENCHMARK PROFILE
iOSWorld
iOSWorld evaluates phone agents on interactive tasks within a native iOS simulator. It includes 26 custom iOS apps with connected personal data, 133 tasks across single-app, multi-app, and memory/personalization categories. Agents are evaluated with rubric-based scoring.
- Released
- 2026-06-08
- Readiness
- Runnable
- Primary field
- Consumer & Productivity
Why it matters
Existing mobile benchmarks lack personalization and interactive evaluation. iOSWorld provides a benchmark with persistent user identity and multi-app tasks, showing significant gaps in multi-app and memory-based performance.
Motivation
A useful phone agent needs to be personally intelligent.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.