WindowsWorld: leaderboard

GUI agents on WindowsWorld (2026), 181 persona-driven professional tasks across 17 Windows desktop applications, mostly multi-application workflows plus infeasible tasks to reject, with the model observing raw screenshots and acting through PyAutoGUI. Share of tasks whose final state is judged correct, in percent; higher is better.

Metric: Final task success rate (%) over all 181 tasks of WindowsWorld (persona-driven Windows desktop tasks across 17 applications, executed with a unified PyAutoGUI action space and level-specific step budgets of 15, 25, 40 and 20), the model observing raw screenshots, states judged by a human-validated Qwen3-VL-Plus judge; infeasible Level 4 tasks count as solved when the agent correctly rejects them; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash (Preview)17.13#152
2Gemini 3 Pro (Preview)8.29#119
3GPT-5.21.78#185
4Qwen 3 VL Plus0.55#512
5Claude Sonnet 4.50#261

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=windowsworld · How It Works · Data refreshed daily, snapshot 2026-11-21.