OSWorld-Verified
Agents · 2025-07-28
Computer-use benchmark on a live Ubuntu VM, scored by binary task-completion success. Reporting protocols vary by source: some average repeated trials, while others report a single attempt or first-attempt success, so trial details belong in each observation's notes.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Mythos Preview | 85.4 |
| Claude Fable 5 | 85.0 |
| Claude Mythos 5 | 85.0 |
| Opus 4.7 | 83.6 |
| Opus 4.8 | 83.4 |
| Gemini 3.6 Flash | 83.0 |
| Sonnet 4.6 | 81.5 |
| Sonnet 5 | 81.2 |
| Muse Spark 1.1 | 80.7 |
| GPT-5.5 | 78.7 |
| Gemini 3.5 Flash | 78.4 |
| Gemini 3.1 Pro Preview | 78.3 |
| MiniMax M3 | 75.2 |
| GPT-5.4 | 75.0 |
| Gemini 3.5 Flash-Lite | 74.0 |
| Qwen3.7-Plus | 73.3 |
| Kimi K2.6 | 73.1 |
| Opus 4.6 | 72.7 |
| GPT-5.6 Luna | 72.6 |
| Opus 4.5 | 72.6 |
| GPT-5.4 Mini | 72.1 |
| GPT-5 | 69.9 |
| Gemini 3 Flash Preview | 65.1 |
| Kimi K2.5 | 63.3 |
| Sonnet 4.5 | 62.9 |
| Doubao Seed 1.8 | 61.9 |
| Gemini 3.1 Flash-Lite | 54.3 |
| Haiku 4.5 | 50.7 |
| Gemini 2.5 Pro | 45.8 |
| Sonnet 4 | 43.9 |
| GPT-5 Mini | 42.0 |
| GPT-5.4 Nano | 39.0 |
| GPT-5.2-Codex | 38.2 |
| Claude 3.7 Sonnet | 35.8 |
| Computer Use Preview | 31.3 |
| GPT-4o | 29.3 |
| UI-TARS 72B DPO | 27.1 |
| o3 | 23.0 |
| Kimi-VL-A3B-Thinking | 10.3 |
| Qwen2.5-VL-72B-Instruct | 5.0 |