# OSWorld (OSWorld-Verified)

Gives an AI control of a real computer (mouse, keyboard, screen) and asks it to do everyday jobs in apps like a browser, spreadsheet or image editor. It shows how close AI is to operating a PC the way a person does.

Category: agents. Type: leaderboard. Status: active.
Metric: Share of the 369 desktop tasks the agent completed successfully (%), averaged over runs where several were submitted (higher is better).
Source: OSWorld team (XLANG Lab, University of Hong Kong) — https://os-world.github.io/static/data/osworld_verified_results.xlsx, results as of 2026-08-01, retrieved 2026-11-17, license: not stated.

| Rank | Model / agent | Organization | Approach type | Max steps | Success rate (%) | Date |
|---|---|---|---|---|---|---|
| 1 | Intelligence-Indeed Agent | Intelligence Indeed | Agentic framework | 100 | 90.2 | 2026-07-25 |
| 2 | claude-fable-5[1m] | Anthropic | General model | 100 | 86.0 | 2026-08-01 |
| 3 | Pointer Agent w/ Opus 4.7 | Pointer | Agentic framework | 100 | 83.6 | 2026-05-21 |
| 4 | claude-opus-5[1m] | Anthropic | General model | 100 | 83.4 | 2026-08-01 |
| 5 | Coasty CUA v1 | Coasty Team | General model | 100 | 82.8 | 2026-07-01 |
| 6 | Pointer Agent w/ Sonnet 4.6 | Pointer | Agentic framework | 100 | 81.5 | 2026-05-21 |
| 7 | Muse Spark 1.1 | Meta Superintelligence Labs | General model | 100 | 80.7 | 2026-07-09 |
| 8 | Holo3-35B-A3B | H Company | Specialized model | 100 | 80.4±2.2 | 2026-04-20 |
| 9 | OpenAPA w/ gemini-3.1-pro | Laiye | Agentic framework | 100 | 78.3 | 2026-04-17 |
| 10 | VLAA-GUI w/ Opus 4.5 | UCSC VLAA Lab | Agentic framework | 100 | 76.3 | 2026-04-20 |
| 11 | MiniMax M3 | MiniMax | General model | 100 | 75.2 | 2026-06-08 |
| 12 | HIPPO Agent w/ Opus 4.5 | Lenovo | Agentic framework | 100 | 74.5 | 2026-02-25 |
| 13 | Qwen 3.7 Plus | Qwen Team, Alibaba Group | General model | 100 | 73.3 | 2026-05-25 |
| 14 | Kimi K2.6 | Moonshot AI | General model | 100 | 73.1 | 2026-04-20 |
| 15 | agent s3 w/ Opus 4.5 + GPT-5 bBoN (N=10) | Simular | Agentic framework | 100 | 72.6 | 2025-12-11 |
| 16 | claude-sonnet-4-6 | Anthropic | General model | 100 | 72.1 | 2026-03-08 |
| 17 | agent s3 w/ GPT-5 bBoN (N=10) | Simular | Agentic framework | 100 | 69.9 | 2025-10-04 |
| 18 | UiPath Screen Agent w/ Opus 4.5 | UiPath | Agentic framework | 100 | 67.1 | 2025-12-24 |
| 19 | agent s3 w/ Opus 4.5 bBoN (N=1) | Simular | Agentic framework | 100 | 66.0±1.1 | 2025-12-11 |
| 20 | OS-Symphony w/ GPT-5 | Shanghai AI Laboratory | Agentic framework | 50 | 65.8 | 2026-01-04 |
| 21 | UiPath Screen Agent w/ Opus 4.5 | UiPath | Agentic framework | 50 | 64.4 | 2025-12-24 |
| 22 | GBOX Agent | GBOX.AI | Agentic framework | 15 | 64.2 | 2025-11-25 |
| 23 | GTA1 w/ GPT-5 | Salesforce & The Australian National University & The University of Hong Kong | Agentic framework | 100 | 63.4 | 2025-10-03 |
| 24 | Kimi K2.5 | Moonshot AI | General model | 100 | 63.3 | 2026-01-31 |
| 25 | claude-sonnet-4-5-20250929 | Anthropic | General model | 100 | 62.9 | 2025-10-31 |

Notes: OSWorld-Verified results file behind the official leaderboard. Rows are grouped by model + max-steps setting and averaged the same way the official page does (mean, with ± population std when several runs exist), sorted by score; all approach types (general models, specialized models, agentic frameworks) included. as_of = latest result date in the file.
