OSWorld
在真实计算机环境中执行开放式任务，按执行结果判定成功；收录官方 OSWorld-Verified 结果与模型方自报结果。
收录成绩204133 个模型或系统
当前第一90.2Intelligence-Indeed Agent
指标任务成功率成功完成的任务占比（%），按执行结果自动判定
最近同步10-16 16:572 个数据源 · 任务执行
排名表
显示 204 / 204
| 应用领域 | |||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 模型 | 机构 | 参数 | 开源 | 评测设置 | 来源 | 备注 | |||||||||||||
| 1 | Intelligence-Indeed Agent | Intelligence Indeed | — | — | 90.2 | 78.1 | 92.3 | 97.9 | 91.4 | 95.7 | 84.7 | 100.0 | 86.7 | 99.3 | 95.7 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 2 | claude-fable-5[1m] | Anthropic | — | — | 86.0 | 78.1 | 88.5 | 95.7 | 80.6 | 95.7 | 79.3 | 95.8 | 80.0 | 86.7 | 100.0 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 3 | Pointer Agent w/ Opus 4.7 | Pointer | — | — | 83.6 | 78.2 | 80.8 | 91.5 | 91.4 | 91.3 | 74.4 | 95.8 | 66.7 | 87.4 | 91.3 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 4 | claude-opus-5[1m] | Anthropic | — | — | 83.4 | 80.2 | 88.0 | 95.7 | 80.6 | 95.7 | 76.8 | 79.2 | 73.3 | 82.3 | 91.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 5 | Coasty CUA v1 | Coasty Team | — | — | 82.8 | 67.3 | 96.2 | 89.4 | 95.7 | 78.3 | 69.2 | 100.0 | 93.3 | 79.9 | 95.7 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 6 | Pointer Agent w/ Sonnet 4.6 | Pointer | — | — | 81.5 | 73.2 | 84.6 | 89.4 | 93.5 | 91.3 | 65.0 | 91.7 | 73.3 | 93.3 | 95.7 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 7 | Muse Spark 1.1 | Meta Superintelligence Labs | — | — | 80.7 | 73.8 | 80.8 | 93.6 | 74.4 | 86.8 | 71.7 | 95.8 | 80.0 | 92.4 | 87.0 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 8 | Holo3-35B-A3B | H Company | — | — | 80.4 | 78.3 | 88.5 | 83.0 | 83.1 | 84.7 | 62.9 | 95.8 | 93.3 | 94.1 | 95.7 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 2.20） |
| 9 | OpenAPA w/ gemini-3.1-pro | Laiye | — | — | 78.3 | 71.7 | 84.6 | 93.6 | 82.9 | 73.9 | 65.7 | 83.3 | 80.0 | 92.7 | 82.6 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 10 | VLAA-GUI w/ Opus 4.5 | UCSC VLAA Lab | — | — | 76.3 | 66.6 | 76.9 | 89.4 | 75.7 | 91.3 | 61.1 | 91.7 | 80.0 | 84.9 | 91.3 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 11 | MiniMax M3 | MiniMax | — | — | 75.2 | 76.0 | 76.9 | 80.8 | 73.7 | 91.2 | 61.6 | 83.3 | 86.7 | 86.6 | 78.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 12 | HIPPO Agent w/ Opus 4.5 | Lenovo | — | — | 74.5 | 60.4 | 96.2 | 83.0 | 68.0 | 69.3 | 64.3 | 87.5 | 73.3 | 88.2 | 95.7 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 13 | Qwen 3.7 Plus | Qwen Team, Alibaba Group | — | — | 73.3 | 71.1 | 65.4 | 78.7 | 85.1 | 87.0 | 59.3 | 79.2 | 73.3 | 82.3 | 82.6 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 14 | Kimi K2.6 | Moonshot AI | — | — | 73.1 | 76.7 | 73.1 | 80.8 | 82.2 | 73.9 | 55.0 | 79.2 | — | 75.7 | 91.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 15 | agent s3 w/ Opus 4.5 + GPT-5 bBoN (N=10) | Simular | — | — | 72.6 | 63.0 | 73.1 | 87.2 | 72.3 | 86.8 | 63.9 | 79.2 | 73.3 | 62.9 | 82.6 | 最多 100 步；额外使用代码动作；多次采样 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 16 | claude-sonnet-4-6 | Anthropic | — | — | 72.1 | 78.5 | 69.2 | 74.5 | 70.2 | 86.8 | 60.2 | 91.7 | 66.7 | 81.4 | 72.7 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 17 | agent s3 w/ GPT-5 bBoN (N=10) | Simular | — | — | 69.9 | 69.5 | 69.2 | 87.2 | 63.7 | 78.1 | 58.7 | 79.2 | 80.0 | 52.1 | 82.6 | 最多 100 步；额外使用代码动作；多次采样 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 18 | UiPath Screen Agent w/ Opus 4.5 | UiPath | — | — | 67.1 | 69.5 | 69.2 | 80.4 | 70.2 | 69.6 | 53.0 | 70.8 | 73.3 | 60.8 | 78.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 19 | AskUI VisionAgent | AskUI | — | — | 66.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | General Model, Max Steps 100, Runs: 1 |
| 20 | agent s3 w/ Opus 4.5 bBoN (N=1) | Simular | — | — | 66.0 | 61.6 | 48.5 | 83.4 | 65.1 | 80.8 | 57.4 | 75.0 | 66.7 | 67.7 | 69.6 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework；5 次运行的平均值（标准差 1.10） |
| 21 | OS-Symphony w/ GPT-5 | Shanghai AI Laboratory | — | — | 65.8 | 63.0 | 80.8 | 76.6 | 48.8 | 78.1 | 55.3 | 78.3 | 80.0 | 61.5 | 78.3 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 22 | UiPath Screen Agent w/ Opus 4.5 | UiPath | — | — | 64.4 | 68.1 | 57.7 | 68.1 | 67.3 | 78.3 | 48.8 | 70.8 | 80.0 | 64.1 | 82.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 23 | GBOX Agent | GBOX.AI | — | — | 64.2 | 63.0 | 76.9 | 74.5 | 63.7 | 68.2 | 49.7 | 70.8 | 66.7 | 46.3 | 91.3 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 24 | GTA1 w/ GPT-5 | Salesforce & The Australian National University & The University of Hong Kong | — | — | 63.4 | 58.6 | 76.9 | 63.8 | 65.5 | 60.7 | 50.9 | 79.2 | 80.0 | 57.9 | 82.6 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 25 | Kimi K2.5 | Moonshot AI | — | — | 63.3 | 63.0 | 69.2 | 68.1 | 65.9 | 78.3 | 46.1 | 73.9 | 80.0 | 63.8 | 78.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 26 | claude-sonnet-4-5-20250929 | Anthropic | — | — | 62.9 | 63.0 | 53.9 | 72.3 | 68.0 | 82.5 | 49.5 | 70.8 | 60.0 | 58.2 | 73.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 27 | agent s3 w/ GPT-5 bBoN (N=1) | Simular | — | — | 62.6 | 58.8 | 64.2 | 78.5 | 53.6 | 67.3 | 51.4 | 77.5 | 74.0 | 57.0 | 76.1 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework；10 次运行的平均值（标准差 1.48） |
| 28 | Agentic-Lybic-Maestro | Lybic | — | — | 61.9 | 63.0 | 73.1 | 60.9 | 69.4 | 65.2 | 41.6 | 75.0 | 66.7 | 80.7 | 82.6 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 29 | Seed-1.8 | ByteDance Seed | — | — | 61.9 | 64.4 | 69.2 | 70.2 | 60.9 | 82.1 | 42.4 | 66.7 | 80.0 | 62.8 | 73.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 30 | CoACT-1 | Salesforce & University of Southern California & University of Washington | — | — | 60.8 | 54.3 | 65.4 | 70.2 | 50.3 | 73.9 | 47.9 | 75.0 | 73.3 | 71.9 | 78.3 | 最多 150 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 31 | CoACT-1 | Salesforce & University of Southern California & University of Washington | — | — | 59.9 | 52.1 | 65.4 | 70.2 | 50.3 | 73.9 | 47.9 | 75.0 | 66.7 | 66.1 | 78.3 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 32 | claude-sonnet-4-5-20250929 | Anthropic | — | — | 58.1 | 56.4 | 57.7 | 66.0 | 57.5 | 65.2 | 47.0 | 70.8 | 66.7 | 52.9 | 69.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 33 | aworldGUIAgent-v1 | inclusionAI | — | — | 58.0 | 49.9 | 73.1 | 70.2 | 60.4 | 56.5 | 43.5 | 66.7 | 73.3 | 57.6 | 69.6 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 34 | Turix (computer_13, 50 steps) | New Generation Technology | — | — | 58.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | Provider: VMware; Observation type: screenshot; Action space: computer_13; Max steps: 50; Parallel environments: 3; Run date: 2026-03-23; Repository: xlang-ai/OSWorld (fork); Commit: 3afe772dc54e5c47ef6916f0de27f7c2b5c34d2b; Test set: evaluation_examples/test_nogdrive_turix.json (no Google Drive). |
| 35 | Agentic-Lybic-Maestro | Lybic | — | — | 56.9 | 61.9 | 78.8 | 51.1 | 60.5 | 64.3 | 38.5 | 66.7 | 70.0 | 60.8 | 71.7 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework；2 次运行的平均值（标准差 0.17） |
| 36 | EvoCUA-20260105 | Meituan LongCat Team | — | — | 56.7 | 65.1 | 76.9 | 55.3 | 59.5 | 69.6 | 27.9 | 78.3 | 80.0 | 49.4 | 87.0 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 37 | CoACT-1 | Salesforce & University of Southern California & University of Washington | — | — | 56.4 | 45.6 | 61.5 | 68.1 | 46.7 | 73.9 | 42.4 | 70.8 | 66.7 | 66.1 | 78.3 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 38 | agent s2.5 w/ o3 | Simular Research | — | — | 56.0 | 52.1 | 76.9 | 59.6 | 55.3 | 47.8 | 44.1 | 75.0 | 73.3 | 42.0 | 73.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 39 | GUI-Owl-1.5 32B | Tongyi Lab, Alibaba Group | — | — | 55.4 | 66.6 | 76.9 | 48.9 | 63.8 | 60.9 | 27.3 | 76.2 | 86.7 | 37.4 | 87.0 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 40 | agent s2.5 w/ o3 | Simular Research | — | — | 54.2 | 52.1 | 76.9 | 55.3 | 55.3 | 47.8 | 39.5 | 75.0 | 73.3 | 42.0 | 73.9 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 41 | DeepMiner-Mano-72B | Mininglamp Technology | — | — | 53.9 | 54.3 | 88.5 | 57.5 | 61.6 | 78.3 | 24.4 | 66.7 | 66.7 | 35.3 | 78.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 42 | UiPath Screen Agent w/ GPT-5 | UiPath | — | — | 53.6 | 58.6 | 61.5 | 40.4 | 53.1 | 60.7 | 37.3 | 73.9 | 80.0 | 55.8 | 82.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 43 | GTA1 w/ o3 | Salesforce & The Australian National University & The University of Hong Kong | — | — | 53.1 | 54.3 | 61.5 | 59.6 | 46.7 | 60.7 | 38.3 | 62.5 | 86.7 | 53.3 | 60.9 | 最多 100 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 43 | UI-TARS-2-2509 | ByteDance Seed | — | — | 53.1 | 63.0 | 50.0 | 66.0 | 56.4 | 60.9 | 34.1 | 41.7 | 73.3 | 49.9 | 73.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 45 | Jedi-7B w/ o3 | Salesforce & The University of Hong Kong | — | — | 51.0 | 58.7 | 69.2 | 42.5 | 44.6 | 56.5 | 35.3 | 50.0 | 80.0 | 54.4 | 82.6 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 46 | Jedi-7B w/ o3 | Salesforce & The University of Hong Kong | — | — | 50.6 | 57.7 | 80.8 | 40.4 | 44.7 | 65.2 | 35.0 | 54.2 | 80.0 | 58.1 | 56.5 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 47 | EvoCUA | Meituan LongCat Team | — | — | 50.3 | 63.0 | 53.9 | 38.3 | 57.4 | 56.5 | 28.2 | 65.2 | 80.0 | 52.2 | 78.3 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 48 | GTA1 w/ o3 | Salesforce & The Australian National University & The University of Hong Kong | — | — | 48.6 | 34.8 | 73.1 | 40.4 | 44.6 | 60.7 | 37.0 | 58.3 | 86.7 | 35.5 | 82.6 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 49 | autoglm-os-9b-20250925 | Tsinghua University & Zhipu AI | — | — | 48.0 | 39.1 | 61.5 | 58.7 | 29.7 | 52.2 | 27.9 | 75.0 | 80.0 | 64.7 | 82.6 | 最多 50 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 50 | autoglm-os-9b | Tsinghua University & Zhipu AI | — | — | 47.3 | 37.0 | 57.7 | 58.7 | 27.6 | 52.2 | 33.4 | 66.7 | 80.0 | 64.3 | 69.6 | 最多 50 步；额外使用无障碍树；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 51 | autoglm-os-9b-20250925 | Tsinghua University & Zhipu AI | — | — | 46.9 | 34.8 | 65.4 | 51.1 | 29.7 | 52.2 | 28.4 | 79.2 | 73.3 | 69.5 | 78.3 | 最多 15 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 52 | autoglm-os-9b | Tsinghua University & Zhipu AI | — | — | 46.3 | 45.6 | 53.9 | 57.5 | 25.4 | 52.2 | 30.9 | 66.7 | 80.0 | 60.4 | 60.9 | 最多 15 步；额外使用无障碍树；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 53 | EvoCUA-8B-20260105 | Meituan LongCat Team | — | — | 46.1 | 54.3 | 84.6 | 29.8 | 38.2 | 52.2 | 25.6 | 75.0 | 73.3 | 44.8 | 65.2 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 54 | agent s2 w/ gemini-2.5-pro | Simular Research | — | — | 45.8 | 45.6 | 65.4 | 44.7 | 36.1 | 69.6 | 28.7 | 62.5 | 66.7 | 40.5 | 69.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 55 | GTA1 w/ o3 (100 steps) | Salesforce & The Australian National University & The University of Hong Kong | — | — | 45.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 56 | opencua-72b-preview | The University of Hong Kong & Moonshot AI | — | — | 45.0 | 52.1 | 74.4 | 35.5 | 48.2 | 56.5 | 22.2 | 58.3 | 57.8 | 37.3 | 70.6 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 0.94） |
| 57 | opencua-72b-preview | The University of Hong Kong & Moonshot AI | — | — | 44.9 | 58.6 | 76.9 | 31.9 | 44.7 | 60.9 | 22.3 | 45.8 | 53.3 | 41.2 | 73.9 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 58 | claude-4-sonnet-20250514 | Anthropic | — | — | 43.9 | 54.3 | 50.0 | 31.9 | 46.7 | 60.9 | 28.5 | 45.8 | 73.3 | 41.2 | 60.9 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 59 | OpenAI CUA o3 (200 steps) | OpenAI | — | — | 42.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 60 | claude-sonnet-4-5-20250929 | Anthropic | — | — | 42.9 | 52.1 | 50.0 | 36.2 | 48.9 | 60.9 | 18.3 | 75.0 | 66.7 | 28.8 | 60.9 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 61 | UI-TARS-1.5 (100 steps) | ByteDance Seed & Tsinghua University | — | — | 42.5 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | Active parameter >20B? |
| 62 | Jedi-7B w/ o3 | Salesforce & The University of Hong Kong | — | — | 42.4 | 52.1 | 65.4 | 29.8 | 42.5 | 65.2 | 20.4 | 50.0 | 73.3 | 47.1 | 56.5 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 63 | UI-TARS-250705 | ByteDance Seed | — | — | 41.8 | 56.4 | 50.0 | 40.4 | 55.3 | 60.9 | 14.7 | 41.7 | 66.7 | 44.0 | 52.2 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 64 | qwen3-vl-flash-2025-10-25 | Qwen Team, Alibaba Group | — | — | 41.6 | 56.4 | 42.3 | 23.0 | 51.0 | 56.5 | 22.0 | 54.2 | 66.7 | 34.4 | 69.6 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 65 | Agent S2 w/ Gemini 2.5 (50 steps) | Simular Research | — | — | 41.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 65 | claude-4-sonnet-20250514 | Anthropic | — | — | 41.4 | 47.7 | 50.0 | 29.8 | 42.5 | 52.2 | 27.9 | 45.8 | 66.7 | 32.8 | 69.6 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 67 | DART-GUI-7B-0924 | BIGAI & BIT & DataCanvas | — | — | 40.5 | 56.4 | 80.8 | 14.9 | 44.7 | 47.8 | 17.8 | 54.2 | 66.7 | 27.2 | 69.6 | 最多 30 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 68 | DeepMiner-Mano-7B | Mininglamp Technology | — | — | 40.1 | 39.1 | 69.2 | 27.7 | 42.5 | 56.5 | 17.2 | 50.0 | 73.3 | 35.3 | 78.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 69 | doubao-1-5-thinking-vision-pro-250717 | ByteDance Seed | — | — | 40.0 | 56.5 | 50.0 | 34.8 | 48.9 | 56.5 | 15.3 | 39.1 | 73.3 | 35.3 | 56.5 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 70 | CoACT-1 | Salesforce & University of Southern California & University of Washington | — | — | 39.8 | 30.4 | 34.6 | 51.1 | 34.0 | 56.5 | 23.8 | 66.7 | 40.0 | 56.5 | 60.9 | 最多 15 步；额外使用代码动作 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 71 | agent s2.5 w/ o3 | Simular Research | — | — | 39.0 | 32.5 | 57.7 | 40.4 | 44.7 | 43.5 | 17.8 | 70.8 | 66.7 | 34.6 | 56.5 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 71 | opencua-72b-preview | The University of Hong Kong & Moonshot AI | — | — | 39.0 | 54.3 | 80.8 | 27.7 | 40.4 | 56.5 | 13.9 | 50.0 | 46.7 | 23.5 | 60.9 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 73 | mobile-agent-v3 w/ gui-owl-32b | Tongyi Lab, Alibaba Group | — | — | 38.9 | 50.9 | 61.5 | 13.0 | 29.7 | 52.2 | 20.4 | 58.3 | 60.0 | 46.1 | 82.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 74 | OpenAI CUA 4o (200 steps) | OpenAI | — | — | 38.1 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 75 | claude-3-7-sonnet-20250219 | Anthropic | — | — | 35.8 | 52.1 | 38.5 | 31.9 | 36.1 | 43.5 | 17.7 | 50.0 | 53.3 | 23.5 | 56.5 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 76 | claude-3-7-sonnet-20250219 | Anthropic | — | — | 35.6 | 54.3 | 42.3 | 21.3 | 31.8 | 47.8 | 19.6 | 45.8 | 66.7 | 24.9 | 56.5 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 77 | InfantAgent (50 steps) | Evolution intelligence & University of Minnesota | — | — | 35.3 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | Planner: Claude-3.7 Sonnet，Grounder: UI-TARS-1.5-7B |
| 78 | opencua-32b | The University of Hong Kong & Moonshot AI | — | — | 34.8 | 39.8 | 66.7 | 18.4 | 37.6 | 36.2 | 16.2 | 55.1 | 46.7 | 33.3 | 63.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 0.78） |
| 79 | agent s2 w/ gemini-2.5-pro | Simular Research | — | — | 34.6 | 32.5 | 53.9 | 27.7 | 27.6 | 56.5 | 13.9 | 58.3 | 66.7 | 40.5 | 69.6 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 80 | Agent S2 w/ Claude 3.7 (50 steps) | Simular Research | — | — | 34.5 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 81 | opencua-32b | The University of Hong Kong & Moonshot AI | — | — | 34.1 | 43.4 | 69.2 | 13.5 | 38.3 | 40.6 | 14.9 | 53.6 | 53.3 | 25.5 | 55.1 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 0.69） |
| 82 | doubao-1-5-thinking-vision-pro-250428 | ByteDance Seed | — | — | 33.8 | 47.6 | 53.9 | 34.8 | 40.4 | 43.5 | 12.2 | 26.1 | 60.0 | 17.6 | 52.2 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 83 | gui-owl-7b | Tongyi Lab, Alibaba Group | — | — | 32.1 | 41.2 | 65.4 | 17.0 | 19.1 | 52.2 | 9.7 | 50.0 | 66.7 | 29.4 | 65.2 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 84 | doubao-1-5-thinking-vision-pro-250717 | ByteDance Seed | — | — | 31.9 | 47.8 | 50.0 | 25.5 | 36.1 | 43.5 | 6.5 | 33.3 | 66.7 | 35.3 | 47.8 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 85 | computer-use-preview | OpenAI | — | — | 31.3 | 36.9 | 34.6 | 14.9 | 29.7 | 26.1 | 15.8 | 70.8 | 66.7 | 11.8 | 69.6 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 86 | claude-4-sonnet-20250514 | Anthropic | — | — | 31.2 | 36.9 | 46.1 | 17.0 | 36.2 | 43.5 | 9.7 | 37.5 | 66.7 | 38.5 | 60.9 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 87 | computer-use-preview | OpenAI | — | — | 30.5 | 33.6 | 48.1 | 18.1 | 24.5 | 23.9 | 15.6 | 54.2 | 56.7 | 38.2 | 60.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 0.88） |
| 88 | UI-TARS-7B-1.5 + ARPO (w. gold reward) | CUHK, SmartMore, HKUST | — | — | 29.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 89 | TianXi-Action-7B | Lenovo | — | — | 29.8 | 36.8 | 55.8 | 6.4 | 38.3 | 54.3 | 6.6 | 38.2 | 43.3 | 31.9 | 67.4 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 0.57） |
| 90 | opencua-32b | The University of Hong Kong & Moonshot AI | — | — | 29.7 | 40.5 | 55.1 | 13.5 | 30.7 | 39.1 | 9.7 | 52.2 | 44.4 | 25.0 | 52.8 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 1.11） |
| 91 | Jedi-7B w/ gpt-4o | Salesforce & The University of Hong Kong | — | — | 29.3 | 30.4 | 76.9 | 8.5 | 17.0 | 34.8 | 13.8 | 50.0 | 60.0 | 17.6 | 65.2 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 92 | opencua-7b | The University of Hong Kong & Moonshot AI | — | — | 28.2 | 38.6 | 43.6 | 13.2 | 32.6 | 33.3 | 12.1 | 43.5 | 42.2 | 28.3 | 47.1 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 0.46） |
| 93 | doubao-1-5-thinking-vision-pro-250428 | ByteDance Seed | — | — | 27.8 | 44.4 | 46.1 | 13.0 | 31.8 | 39.1 | 4.8 | 30.4 | 66.7 | 23.5 | 56.5 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 94 | uitars-1.5-7b | ByteDance Seed & Tsinghua University | — | — | 27.4 | 38.3 | 51.9 | 9.6 | 38.2 | 39.1 | 8.9 | 31.3 | 40.0 | 22.4 | 47.8 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 2.20） |
| 95 | uitars-1.5-7b | ByteDance Seed & Tsinghua University | — | — | 27.3 | 32.8 | 53.8 | 8.5 | 39.3 | 41.3 | 8.6 | 25.6 | 46.7 | 24.4 | 52.2 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 2.15） |
| 96 | claude-3-7-sonnet-20250219 | Anthropic | — | — | 27.1 | 41.2 | 34.6 | 8.5 | 29.7 | 39.1 | 10.8 | 50.0 | 33.3 | 35.3 | 43.5 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 96 | uitars-72b-dpo | ByteDance Seed & Tsinghua University | — | — | 27.1 | 32.6 | 73.1 | 6.4 | 23.8 | 34.8 | 8.3 | 37.5 | 60.0 | 17.6 | 52.2 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 98 | Jedi-7B w/ GPT-4o (100 steps) | Salesforce & The University of Hong Kong | — | — | 27.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | avg 27.04, std 1.81 |
| 99 | Jedi-7B w/ gpt-4o | Salesforce & The University of Hong Kong | — | — | 27.0 | 32.5 | 61.5 | 6.4 | 16.9 | 30.4 | 15.8 | 45.8 | 46.7 | 26.9 | 47.8 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 99 | Simular Agent S2 (15 steps) | Simular Research | — | — | 27.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 101 | UI-TARS-1.5 7B (100 steps) | ByteDance Seed & Tsinghua University | — | — | 26.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 102 | Jedi-7B w/ gpt-4o | Salesforce & The University of Hong Kong | — | — | 26.8 | 31.0 | 69.2 | 6.4 | 14.8 | 47.8 | 8.7 | 45.8 | 53.3 | 26.9 | 52.2 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Agentic framework |
| 103 | opencua-7b | The University of Hong Kong & Moonshot AI | — | — | 26.6 | 36.1 | 47.4 | 10.6 | 31.9 | 27.5 | 9.9 | 42.9 | 40.0 | 29.4 | 44.9 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 0.45） |
| 104 | computer-use-preview | OpenAI | — | — | 26.0 | 36.9 | 34.6 | 10.6 | 25.4 | 30.4 | 10.8 | 45.8 | 46.7 | 29.4 | 47.8 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 105 | uitars-72b-dpo | ByteDance Seed & Tsinghua University | — | — | 25.8 | 33.2 | 61.5 | 12.8 | 25.4 | 43.5 | 6.7 | 33.3 | 33.3 | 23.5 | 47.8 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 106 | Jedi-7B w/ GPT-4o (50 steps) | Salesforce & The University of Hong Kong | — | — | 25.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | avg 25.02, std 1.30 |
| 107 | UI-TARS-72B-DPO (50 steps) | ByteDance Seed & Tsinghua University | — | — | 24.6 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 108 | OSCAR w/ GPT-4o (w. gold reward) | Universite de Montreal & Mila | — | — | 24.5 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | |
| 108 | uitars-1.5-7b | ByteDance Seed & Tsinghua University | — | — | 24.5 | 22.9 | 51.9 | 11.7 | 29.7 | 39.1 | 3.8 | 34.8 | 26.7 | 34.2 | 63.0 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；2 次运行的平均值（标准差 1.20） |
| 110 | opencua-7b | The University of Hong Kong & Moonshot AI | — | — | 24.3 | 32.5 | 44.9 | 9.2 | 30.5 | 26.1 | 7.9 | 36.2 | 44.4 | 23.5 | 43.5 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model；3 次运行的平均值（标准差 1.39） |
| 111 | uitars-72b-dpo | ByteDance Seed & Tsinghua University | — | — | 24.0 | 37.6 | 61.5 | 8.7 | 25.5 | 30.4 | 4.9 | 33.3 | 40.0 | 17.6 | 39.1 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 112 | AgentStore (AgentToken) w/ GPT-4o | Shanghai AI Lab | — | — | 23.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | |
| 113 | opencua-qwen2-7b | The University of Hong Kong & Moonshot AI | — | — | 23.1 | 26.0 | 38.5 | 10.6 | 29.7 | 43.5 | 6.6 | 25.0 | 33.3 | 23.5 | 47.8 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 114 | o3 | OpenAI | — | — | 23.0 | 13.0 | 38.5 | 10.6 | 10.6 | 30.4 | 16.5 | 62.5 | 26.7 | 39.2 | 39.1 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 115 | Jedi-7B w/ GPT-4o (15 steps) | Salesforce & The University of Hong Kong | — | — | 22.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | avg 22.74, std 0.51 |
| 116 | UI-TARS-72B-DPO (15 steps) | ByteDance Seed & Tsinghua University | — | — | 22.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 117 | learn-by-interact w/ Claude-3.5-sonnet | Google Cloud & The University of Hong Kong | — | — | 22.5 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=0 |
| 118 | Claude 3.5 Sonnet (50 steps) | Anthropic | — | — | 22.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 119 | Jedi-3B w/ GPT-4o (15 steps) | Salesforce & The University of Hong Kong | — | — | 21.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | avg 21.8, std 0.19 |
| 120 | opencua-qwen2-7b | The University of Hong Kong & Moonshot AI | — | — | 20.6 | 32.5 | 23.1 | 6.4 | 25.5 | 43.5 | 4.3 | 8.7 | 46.7 | 29.4 | 43.5 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 121 | Agent S w/ GPT-4o | Simular Research | — | — | 20.6 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 122 | Agent S w/ Claude-3.5 | Simular Research | — | — | 20.5 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 123 | UI-TARS-7B-DPO + ZeroGUI | Shanghai AI Lab | — | — | 20.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 124 | opencua-a3b | The University of Hong Kong & Moonshot AI | — | — | 19.9 | 21.6 | 53.9 | 2.1 | 23.3 | 34.8 | 6.5 | 37.5 | 6.7 | 11.8 | 43.5 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 124 | opencua-qwen2-7b | The University of Hong Kong & Moonshot AI | — | — | 19.9 | 19.5 | 38.5 | 8.5 | 23.4 | 34.8 | 3.2 | 16.7 | 40.0 | 29.4 | 52.2 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 126 | OpenAI CUA 4o (15 steps) | OpenAI | — | — | 19.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 127 | R-MCTS w/ GPT-4o (max 60 tree nodes) | Columbia University & Microsoft Research | — | — | 19.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | |
| 128 | UI-TARS-72B-SFT (15 steps) | ByteDance Seed & Tsinghua University | — | — | 18.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 129 | UI-TARS-7B-DPO (15 steps) | ByteDance Seed & Tsinghua University | — | — | 18.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 130 | opencua-a3b | The University of Hong Kong & Moonshot AI | — | — | 17.7 | 26.0 | 53.9 | 6.4 | 25.7 | 17.4 | 2.1 | 12.5 | 20.0 | 11.8 | 39.1 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 130 | UI-TARS-7B-SFT (15 steps) | ByteDance Seed & Tsinghua University | — | — | 17.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 132 | AgentStore (Fine-Tuning) w/ GPT-4o | Shanghai AI Lab | — | — | 17.3 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | |
| 133 | o3 | OpenAI | — | — | 17.2 | 21.7 | 38.5 | 8.5 | 4.3 | 21.7 | 11.8 | 37.5 | 20.0 | 17.6 | 21.7 | 最多 50 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 134 | Aguvis-72B w/ GPT-4o | Salesforce & The University of Hong Kong | — | — | 17.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 135 | opencua-a3b | The University of Hong Kong & Moonshot AI | — | — | 16.9 | 25.9 | 38.5 | 2.1 | 19.1 | 21.7 | 5.4 | 16.7 | 20.0 | 23.5 | 34.8 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：Specialized model |
| 136 | R-MCTS w/ GPT-4o (max 60 tree nodes) | Columbia University & Microsoft Research | — | — | 16.6 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | |
| 137 | Aria-UI w/ GPT-4o | University of Hong Kong & Rhymes AI | — | — | 15.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 138 | Claude 3.5 Sonnet (15 steps) | Anthropic | — | — | 14.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 138 | PC Agent-E | GAIR, Shanghai Jiao Tong University | — | — | 14.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 140 | Aguvis-7B w/ GPT-4o | Salesforce & The University of Hong Kong | — | — | 14.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 141 | OS-Atlas-Base-7B w/ GPT-4o | Shanghai AI Lab | — | — | 14.6 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 142 | AgentStore (In-Context Learning) w/ GPT-4o | Shanghai AI Lab | — | — | 13.6 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | |
| 143 | GPT-4 | OpenAI | — | — | 12.2 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 144 | GPT-4 Vision | OpenAI | — | — | 12.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 145 | GPT-4 Vision | OpenAI | — | — | 11.8 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 146 | OS-Atlas-Base-4B w/ GPT-4o | Shanghai AI Lab | — | — | 11.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 147 | GPT-4o | OpenAI | — | — | 11.4 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 148 | GPT-4o | OpenAI | — | — | 11.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 149 | Friday | Shanghai AI Lab | — | — | 11.1 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 150 | GPT-4 Vision (0409) | OpenAI | — | — | 10.8 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 151 | kimi-vl-a3b | Moonshot AI | — | — | 10.3 | 23.7 | 12.0 | 2.1 | 6.5 | 13.0 | 1.1 | 12.5 | 20.0 | 17.6 | 26.1 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 151 | learn-by-interact w. Gemini-1.5-pro | Google Cloud & The University of Hong Kong authors | — | — | 10.3 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=0 |
| 153 | Aguvis-72B | Salesforce & The University of Hong Kong | — | — | 10.3 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 154 | kimi-vl-a3b | Moonshot AI | — | — | 9.7 | 24.4 | 15.4 | 4.3 | 10.6 | 13.0 | 1.1 | 0.0 | 20.0 | 17.6 | 13.0 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 155 | SeeClick w/ GPT-4o | Nanjing University & Shanghai AI Lab | — | — | 9.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 156 | o3 | OpenAI | — | — | 9.1 | 6.5 | 11.5 | 0.0 | 0.0 | 4.3 | 11.8 | 37.5 | 6.7 | 10.9 | 13.0 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 157 | GPT-4 Vision (0409) | OpenAI | — | — | 9.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 158 | Open-Interpreter | openinterpreter | — | — | 8.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 159 | Qwen2.5-vl-72B | Alibaba Cloud, Qwen Team | — | — | 8.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 160 | Ponder&Press w/ GPT-4o | Shenzhen International Graduate School, Tsinghua University | — | — | 8.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 161 | GPT-4 Vision (0409) | OpenAI | — | — | 8.4 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 162 | Kimi-VL | Kimi & The University of Hong Kong | — | — | 8.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | |
| 163 | CogAgent-9B-20241220 | Tsinghua University & Zhipu AI | — | — | 8.1 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 164 | CRADLE w/ GPT-4o | BAAI | — | — | 7.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 165 | Gemini-Pro-1.5 | — | — | 7.8 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 166 | GPT-4 Vision | OpenAI | — | — | 7.7 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 167 | Qwen-Max | Alibaba Cloud, Qwen Team | — | — | 6.9 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 168 | Claude-3-Opus | AnthropicAI | — | — | 6.7 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 169 | GPT-4o-mini | OpenAI | — | — | 6.2 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 169 | GPT-4o-mini | OpenAI | — | — | 6.2 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 171 | Gemini-Pro Vision | — | — | 5.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 172 | Gemini-Pro-1.5 | — | — | 5.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 172 | GPT-4 Vision (0409) | OpenAI | — | — | 5.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 172 | Qwen-vl-Max-0809 | Alibaba Cloud, Qwen Team | — | — | 5.4 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 175 | Gemini-Pro-1.5 | — | — | 5.1 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 176 | GPT-4o | OpenAI | — | — | 5.0 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 177 | qwen2.5-vl-72b-instruct | Alibaba Cloud, Qwen Team | — | — | 5.0 | 4.3 | 0.0 | 6.4 | 0.0 | 8.7 | 3.2 | 16.7 | 13.3 | 5.9 | 4.3 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 178 | Gemini-Pro-1.5 | — | — | 4.8 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 179 | GPT-4o | OpenAI | — | — | 4.6 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 180 | qwen2.5-vl-72b-instruct | Alibaba Cloud, Qwen Team | — | — | 4.4 | 10.9 | 0.0 | 4.3 | 0.0 | 4.3 | 5.4 | 8.3 | 0.0 | 5.9 | 0.0 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 181 | Claude-3-Opus | AnthropicAI | — | — | 4.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 182 | Qwen-vl-Max-0809 | Qwen | — | — | 4.3 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 183 | qwen2.5-vl-32b-instruct | Alibaba Cloud, Qwen Team | — | — | 3.9 | 8.7 | 3.9 | 0.0 | 0.0 | 8.7 | 2.1 | 8.3 | 6.7 | 0.0 | 8.7 | 最多 100 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 184 | GPT-4o-mini | OpenAI | — | — | 3.8 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 185 | GPT-4o-mini | OpenAI | — | — | 3.5 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 185 | Qwen-vl-Max-0809 | Alibaba Cloud, Qwen Team | — | — | 3.5 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 187 | Gemini-Pro Vision | — | — | 3.5 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 188 | InternVL2 | OpenGVLab | — | — | 3.3 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 189 | qwen2.5-vl-32b-instruct | Alibaba Cloud, Qwen Team | — | — | 3.0 | 8.7 | 3.9 | 0.0 | 0.0 | 8.7 | 0.0 | 8.3 | 13.3 | 0.0 | 0.0 | 最多 15 步 | 官方榜单 | 2026-10-16 | 方法类型：General model |
| 190 | Mixtral-8x7B | MistralAI | — | — | 3.0 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 191 | Llava-OneVision | ByteDance & NTU & CUHK & HKUST | — | — | 2.7 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 192 | GPT-3.5 | OpenAI | — | — | 2.7 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 193 | Claude-3-Opus | AnthropicAI | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 193 | Llava-OneVision | ByteDance & NTU & CUHK & HKUST | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 193 | Llava-OneVision | ByteDance & NTU & CUHK & HKUST | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 193 | Llava-OneVision | ByteDance & NTU & CUHK & HKUST | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 193 | Qwen-vl-Max-0809 | Alibaba Cloud, Qwen Team | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 198 | Gemini-Pro | — | — | 2.4 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 199 | MiniCPM-V 2.6 | MiniCPM-V Team & OpenBMB | — | — | 1.9 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 200 | Llama-3-70B | Meta | — | — | 1.6 | — | — | — | — | — | — | — | — | — | — | 观测：无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 201 | CogAgent | Tsinghua University & Zhipu AI | — | — | 1.3 | — | — | — | — | — | — | — | — | — | — | 观测：截图 + 无障碍树 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 202 | CogAgent | Tsinghua University & Zhipu AI | — | — | 1.1 | — | — | — | — | — | — | — | — | — | — | 观测：截图 | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
| 203 | Gemini-Pro Vision | — | — | 1.1 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 | |
| 204 | CogAgent | Tsinghua University & Zhipu AI | — | — | 1.0 | — | — | — | — | — | — | — | — | — | — | 观测：Set-of-Mark | 模型方自报 | 2026-10-16 | t=1.0, top-p=0.9 |
榜单说明
第 1 批接入- 指标
- 任务成功率：成功完成的任务占比（%），按执行结果自动判定（越高越好）
- 子项
- 应用领域ChromeGIMPCalcImpressWriter跨应用操作系统ThunderbirdVLCVS Code
- 数据源
- 来源类型
- 官方榜单 110模型方自报 94
分数取自数据源原文；同一模型在不同评测设置或不同来源下分别列出，来源冲突时全部保留并标注来源类型。
更新记录
0 条首次同步作为基线，之后新上榜、分数变化和移除的条目会记录在这里。