MacAgentBench: leaderboard

Metric: Pass@1 (%; share of the 676 macOS desktop tasks in 25 applications solved on a single attempt, judged by deterministic rule-based state checks; the benchmark's baseline pure-GUI agent (screenshots and mouse/keyboard actions only), at most 50 steps). Source: arxiv.org. Saturation forecast: Around May 2027. 16 models tracked.

Top models

#ModelScore
1GPT-5.458.4
2Claude Opus 4.639.2
3Gemini 3.1 Pro (Preview)34.2
4Qwen 3 VL 235B A22B21.6
5Qwen 3 VL 32B (Thinking)21.3
6Qwen 3 VL 8B (Thinking)14.5
7uitars-1.5-7B9.8

Interactive version: theaggregate.ai/benchmark?slug=macagentbench · How It Works · Data refreshed daily, snapshot 2026-09-26.