iOSWorld (Vision+XML): leaderboard

Metric: Task pass rate (%) over all 133 tasks with the accessibility tree added to vision (vision+XML), GPT-5.4-Mini judge, 50-step budget; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 6 models tracked.

Top models

#ModelScore
1Claude Opus 4.651.9
2Claude Sonnet 4.646.6
3GPT-5.439.8
4Gemini 3 Flash28.6
5GPT-5.4 Mini15.8
6Qwen 3.5 35B A3B10.5

Interactive version: theaggregate.ai/benchmark?slug=iosworld-vision-plusxml · How It Works · Data refreshed daily, snapshot 2026-09-29.