OSWorld-Verified: leaderboard
369 computer-use tasks in a live desktop OS (LibreOffice, GIMP, VS Code, Chrome) from XLANG Lab at HKU; the 2025 Verified revision fixed 300+ task issues. Execution-checked success rate, humans 72%.
Metric: Success rate (self-reported). Source: benchmarklist.com. Status: years away from saturation. 131 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.8 Max | 86.1 |
| 2 | Claude Fable 5 (Max) | 85 |
| 3 | Qwen 3.8 27B | 84.3 |
| 4 | Claude Opus 4.8 | 83.4 |
| 5 | Gemini 3.6 Flash | 83 |
| 6 | Claude Opus 4.7 | 82.8 |
| 7 | Claude Sonnet 5 | 81.2 |
| 8 | Muse Spark 1.1 | 80.8 |
| 9 | Claude Mythos Preview | 79.6 |
| 10 | GPT-5.5 | 78.7 |
| 11 | Gemini 3.5 Flash | 78.4 |
| 12 | Gemini 3.1 Pro (Preview) | 76.2 |
| 13 | GPT-5.4 (xHigh) | 75 |
| 14 | Gemini 3.5 Flash Lite | 74 |
| 15 | Qwen 3.7 Plus | 73.3 |
Interactive version: theaggregate.ai/benchmark?slug=osworld-verified · How It Works · Data refreshed daily, snapshot 2026-09-05.