DeskCraft (Standard): leaderboard
DeskCraft is an execution-based desktop GUI-agent benchmark of 538 professional-software tasks (386 standard, 152 interactive) across eleven applications, verified from the resulting desktop state. This board reports Task success rate (%) on the 386-task standard split, verified from the final desktop state; higher is better.
Metric: Task success rate (%) on the 386-task standard split, verified from the final desktop state; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.6 | 33.8 | #178 |
| 2 | GPT-5.4 | 31.6 | #123 |
| 3 | Kimi K2.5 | 20.3 | #256 |
| 4 | Qwen 3.5 397B A17B | 12.9 | #246 |
| 5 | Qwen 3.5 35B A3B | 11.1 | #482 |
| 6 | Qwen 3.6 35B A3B | 10.2 | #462 |
| 7 | Qwen 3.5 9B | 8.7 | #803 |
| 8 | Qwen 3 VL 235B A22B | 4.3 | #479 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=deskcraft-standard · How It Works · Data refreshed daily, snapshot 2026-11-13.