OSWorld (self-reported, Screenshot + A11y tree)
Agents · 2024-04-11
Self-reported OSWorld success rate with screenshot and accessibility-tree observation.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-4o | 23.9 |
| Claude 3.5 Sonnet (June 2024) | 20.5 |
| GPT-4 Turbo with Vision | 12.2 |
| GPT-4 Turbo | 9.0 |
| GPT-4o Mini | 6.2 |
| Gemini 1.5 Pro | 5.1 |
| Opus 3 | 4.4 |
| Qwen VL Max 0809 | 3.5 |
| Gemini 1.0 Pro Vision | 3.5 |
| LLaVA-OneVision 72B | 2.4 |
| CogAgent | 1.3 |