Last updated Aug 31, 2026
OSWorld-Verified Leaderboard
Tracking reported computer-use agent performance on OSWorld-Verified, combining the official benchmark leaderboard with additional results found by our Scout.
General models selected
| Model | Institution | Approach type | Max steps | Success rate | Runs | Date | A11y tree | Coding action | Multi-rollout |
|---|---|---|---|---|---|---|---|---|---|
| claude-fable-5[1m] | Anthropic | General model | 100 | 86.0% | 1 | Aug 01, 2026 | No | No | No |
| Navigator n2 | Yutori | General model | 100 | 85.3% | 3 | Aug 26, 2026 | No | Yes | No |
| Kimi K3 | Moonshot AI | General model | – | 84.8% | – | – | – | – | – |
| Claude Opus 4.8 | Anthropic | General model | – | 83.4% | – | – | – | – | – |
| claude-opus-5[1m] | Anthropic | General model | 100 | 83.4% | 1 | Aug 01, 2026 | No | No | No |
| GPT-5.6 Sol | OpenAI | General model | – | 83.0%1 | – | – | – | – | – |
| Gemini 3.6 Flash | General model | – | 83.0% | – | – | – | – | – | |
| Coasty CUA v1 | Coasty Team | General model | 100 | 82.8% | 1 | Jul 01, 2026 | No | Yes | No |
| Claude Sonnet 5 | Anthropic | General model | – | 81.2% | – | – | – | – | – |
| Muse Spark 1.1 | Meta Superintelligence Labs | General model | 100 | 80.7% | 1 | Jul 09, 2026 | No | No | No |
| GPT-5.5 | OpenAI | General model | – | 78.7% | – | – | – | – | – |
| Gemini 3.5 Flash | General model | – | 78.4% | – | – | – | – | – | |
| Claude Opus 4.7 | Anthropic | General model | – | 78.0% | – | – | – | – | – |
| Gemini 3.1 Pro | General model | – | 76.2% | – | – | – | – | – | |
| MiniMax M3 | MiniMax | General model | 100 | 75.2% | 1 | Jun 08, 2026 | No | No | No |
| GPT-5.4 | OpenAI | General model | – | 75.0% | – | – | – | – | – |
| Gemini 3.5 Flash-Lite | General model | – | 74.0% | – | – | – | – | – | |
| Qwen 3.7 Plus | Qwen Team, Alibaba Group | General model | 100 | 73.3% | 1 | May 25, 2026 | No | No | No |
| Kimi K2.6 | Moonshot AI | General model | 100 | 73.1% | 1 | Apr 20, 2026 | No | No | No |
| Claude Opus 4.6 | Anthropic | General model | – | 72.7% | – | – | – | – | – |
| claude-sonnet-4-6 | Anthropic | General model | 100 | 72.1% | 1 | Mar 08, 2026 | No | No | No |
| GPT-5.4 mini | OpenAI | General model | – | 72.1% | – | – | – | – | – |
| Muse Glimmer-30B | Meta | General model | – | 65.9% | – | – | – | – | – |
| Kimi K2.5 | Moonshot AI | General model | 100 | 63.3% | 1 | Jan 31, 2026 | No | No | No |
| claude-sonnet-4-5-20250929 | Anthropic | General model | 100 | 62.9% | 1 | Oct 31, 2025 | No | No | No |
| Seed-1.8 | ByteDance Seed | General model | 100 | 61.9% | 1 | Dec 18, 2025 | No | No | No |
| claude-sonnet-4-5-20250929 | Anthropic | General model | 50 | 58.1% | 1 | Oct 31, 2025 | No | No | No |
| EvoCUA-20260105 | Meituan LongCat Team | General model | 50 | 56.7% | 1 | Jan 06, 2026 | No | No | No |
| EvoCUA | Meituan LongCat Team | General model | 50 | 50.3% | 1 | Dec 24, 2025 | No | No | No |
| EvoCUA-8B-20260105 | Meituan LongCat Team | General model | 50 | 46.1% | 1 | Jan 13, 2026 | No | No | No |
| claude-4-sonnet-20250514 | Anthropic | General model | 50 | 43.9% | 1 | Jul 28, 2025 | No | No | No |
| claude-sonnet-4-5-20250929 | Anthropic | General model | 15 | 42.9% | 1 | Oct 31, 2025 | No | No | No |
| UI-TARS-250705 | ByteDance Seed | General model | 100 | 41.8% | 1 | Sep 16, 2025 | No | No | No |
| qwen3-vl-flash-2025-10-25 | Qwen Team, Alibaba Group | General model | 100 | 41.6% | 1 | Oct 23, 2025 | No | No | No |
| claude-4-sonnet-20250514 | Anthropic | General model | 100 | 41.4% | 1 | Jul 28, 2025 | No | No | No |
| doubao-1-5-thinking-vision-pro-250717 | ByteDance Seed | General model | 100 | 40.0% | 1 | Jul 28, 2025 | No | No | No |
| claude-3-7-sonnet-20250219 | Anthropic | General model | 50 | 35.8% | 1 | Jul 28, 2025 | No | No | No |
| claude-3-7-sonnet-20250219 | Anthropic | General model | 100 | 35.6% | 1 | Jul 28, 2025 | No | No | No |
| doubao-1-5-thinking-vision-pro-250428 | ByteDance Seed | General model | 100 | 33.8% | 1 | Jul 28, 2025 | No | No | No |
| doubao-1-5-thinking-vision-pro-250717 | ByteDance Seed | General model | 15 | 31.9% | 1 | Jul 28, 2025 | No | No | No |
| claude-4-sonnet-20250514 | Anthropic | General model | 15 | 31.2% | 1 | Jul 28, 2025 | No | No | No |
| doubao-1-5-thinking-vision-pro-250428 | ByteDance Seed | General model | 15 | 27.8% | 1 | Jul 28, 2025 | No | No | No |
| claude-3-7-sonnet-20250219 | Anthropic | General model | 15 | 27.1% | 1 | Jul 28, 2025 | No | No | No |
| o3 | OpenAI | General model | 100 | 23.0% | 1 | Jul 28, 2025 | No | No | No |
| o3 | OpenAI | General model | 50 | 17.2% | 1 | Jul 28, 2025 | No | No | No |
| kimi-vl-a3b | Moonshot AI | General model | 100 | 10.3% | 1 | Jul 28, 2025 | No | No | No |
| kimi-vl-a3b | Moonshot AI | General model | 15 | 9.7% | 1 | Jul 28, 2025 | No | No | No |
| o3 | OpenAI | General model | 15 | 9.1% | 1 | Jul 28, 2025 | No | No | No |
| qwen2.5-vl-72b-instruct | Alibaba Cloud, Qwen Team | General model | 100 | 5.0% | 1 | Jul 28, 2025 | No | No | No |
| qwen2.5-vl-72b-instruct | Alibaba Cloud, Qwen Team | General model | 15 | 4.4% | 1 | Jul 28, 2025 | No | No | No |
| qwen2.5-vl-32b-instruct | Alibaba Cloud, Qwen Team | General model | 100 | 3.9% | 1 | Jul 28, 2025 | No | No | No |
| qwen2.5-vl-32b-instruct | Alibaba Cloud, Qwen Team | General model | 15 | 3.0% | 1 | Jul 28, 2025 | No | No | No |
Frequently asked questions
Resources
Paper (arXiv 2404.07972)Code (GitHub · XLANG)Verified trajectories (Hugging Face)Official leaderboardOfficial results workbook (XLSX)Navigator n2 launch post
Citation
@misc{OSWorld,
title={OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments},
author={Tianbao Xie and Danyang Zhang and Jixuan Chen and Xiaochuan Li and Siheng Zhao and Ruisheng Cao and Toh Jing Hua and Zhoujun Cheng and Dongchan Shin and Fangyu Lei and Yitao Liu and Yiheng Xu and Shuyan Zhou and Silvio Savarese and Caiming Xiong and Victor Zhong and Tao Yu},
year={2024},
eprint={2404.07972},
archivePrefix={arXiv},
primaryClass={cs.AI},
}