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. Saturation forecast: Around January 2027. 66 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 85.96 | #16 |
| 2 | Muse Spark 1.1 | 80.67 | #52 |
| 3 | MiniMax-M3 | 75.19 | #136 |
| 4 | Qwen 3.7 Plus | 73.3 | #92 |
| 5 | Kimi K2.6 | 73.06 | #104 |
| 6 | Median Human | 72.36 | |
| 7 | Claude Sonnet 4.6 | 72.11 | #90 |
| 8 | Kimi K2.5 | 63.3 | #144 |
| 9 | Claude Sonnet 4.5 | 62.88 | #142 |
| 10 | Seed 1.8 | 61.87 | #139 |
| 11 | Claude Sonnet 4 (20250514) | 43.9 | #213 |
| 12 | Claude 3.7 Sonnet (20250219) | 35.8 | #201 |
| 13 | GUI-Owl-7B | 32.11 | #852 |
| 14 | uitars-1.5-7B | 29.6 | #864 |
| 15 | O3 | 23 | #124 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=osworld · How It Works · Data refreshed daily, snapshot 2026-10-20.