OSWorld — leaderboard
Computer-use agent benchmark with 369 real-world tasks across Ubuntu, Windows, and macOS. Tests ability to interact with GUIs, terminals, and applications via screenshots and actions.
Metric: Success Rate (%). Source: os-world.github.io. Status: saturation imminent. 63 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Muse Spark 1.1 | 80.67 |
| 2 | MiniMax-M3 | 75.19 |
| 3 | Qwen 3.7 Plus | 73.3 |
| 4 | Kimi K2.6 | 73.06 |
| 5 | Median Human | 72.36 |
| 6 | Claude Sonnet 4.6 | 72.11 |
| 7 | Kimi K2.5 | 63.3 |
| 8 | Claude Sonnet 4.5 | 62.88 |
| 9 | Seed 1.8 | 61.87 |
| 10 | Claude Sonnet 4 (20250514) | 43.9 |
| 11 | Claude 3.7 Sonnet (20250219) | 35.8 |
| 12 | uitars-1.5-7B | 29.6 |
| 13 | O3 | 23 |
| 14 | Kimi-VL-A3B | 10.3 |
| 15 | Qwen 2.5 VL 32B Instruct | 3.88 |
Interactive version: theaggregate.ai/benchmark?slug=osworld · How the rankings work · Data refreshed daily, snapshot 2026-07-22.