OSWorld
Agents · 2024-04-11
A computer-use benchmark where agents complete real desktop and web tasks in reproducible OS environments using keyboard/mouse actions and structured UI observations.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Mythos Preview | 79.6 |
| GPT-5.4 | 75.0 |
| Opus 4.6 | 72.7 |
| Sonnet 4.6 | 72.1 |
| Opus 4.5 | 66.3 |
| o3 | 45.2 |
| Sonnet 4 | 43.9 |
| Gemini 2.5 Pro Preview 03-25 | 41.4 |
| Computer Use Preview | 38.1 |
| Claude 3.7 Sonnet | 35.8 |
| GPT-4o | 27.0 |
| UI-TARS 72B DPO | 24.6 |
| Claude 3.5 Sonnet (Oct 2024) | 22.0 |
| Qwen2.5-VL-72B-Instruct | 8.8 |
| Kimi-VL-A3B-Instruct | 8.2 |
| Gemini 1.0 Pro Vision | 5.8 |
| GPT-4 Turbo | 5.4 |
| Gemini 1.5 Pro | 5.4 |
| GPT-4o Mini | 3.8 |
| Opus 3 | 2.4 |
| LLaVA-OneVision 72B | 2.4 |
| Qwen VL Max 0809 | 2.4 |
| MiniCPM-V 2.6 | 1.9 |
| CogAgent | 1.1 |