MacAgentBench - Internet: leaderboard

Metric: Pass@1 (%; share of the 108 Internet 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 December 2026. 16 models tracked.

Top models

#ModelScore
1GPT-5.451.9
2Claude Opus 4.650
3Gemini 3.1 Pro (Preview)31.5
4Qwen 3 VL 32B (Thinking)17.6
5Qwen 3 VL 235B A22B17.6
6Qwen 3 VL 8B (Thinking)12
7uitars-1.5-7B0.9

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