MacAgentBench - System: leaderboard

Metric: Pass@1 (%; share of the 44 System tasks solved on a single attempt, deterministic rule-based checks; the benchmark's baseline pure-GUI agent (screenshots and mouse/keyboard actions only), at most 50 steps). Source: arxiv.org. Saturation forecast: Around January 2028. 16 models tracked.

Top models

#ModelScore
1GPT-5.452.3
2Claude Opus 4.640.9
3Gemini 3.1 Pro (Preview)34.1
4Qwen 3 VL 235B A22B29.5
5Qwen 3 VL 32B (Thinking)27.3
6Qwen 3 VL 8B (Thinking)18.2
7uitars-1.5-7B13.6

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