OpenAI's benchmark for evaluating AI agents on complex web browsing and information retrieval tasks that require multi-step reasoning over live web content.
Tests real-world information retrieval requiring browsing, reasoning, and synthesis across web pages.
Requires agentic capabilities (browsing). Models without tool use score zero.
Higher is better. Accuracy percentage on retrieval tasks (0–100%).
Requires agentic capabilities (browsing).
| # | Model | Score |
|---|---|---|
| 1 | 99.5 | |
| 2 | 99.2 | |
| 3 | 98.7 | |
| 4 | 93.9 | |
| 5 | 93.7 | |
| 6 | 93.5 | |
| 7 | 91.9 | |
| 8 | 91.6 | |
| 9 | 91.4 | |
| 10 | 91.3 | |
| 11 | 91.0 | |
| 12 | 90.8 | |
| 13 | 90.7 | |
| 14 | 90.2 | |
| 15 | 89.9 | |
| 16 | 89.8 | |
| 17 | 89.5 | |
| 18 | 89.4 | |
| 19 | 89.3 | |
| 20 | 89.2 | |
| 21 | 89.2 | |
| 22 | 88.7 | |
| 23 | 88.7 | |
| 24 | 88.1 | |
| 25 | 87.8 | |
| 26 | 87.8 | |
| 27 | 87.7 | |
| 28 | 87.5 | |
| 29 | 87.5 | |
| 30 | 87.4 | |
| 31 | 87.3 | |
| 32 | 86.9 | |
| 33 | 86.9 | |
| 34 | 86.8 | |
| 35 | 86.7 | |
| 36 | 86.6 | |
| 37 | 85.9 | |
| 38 | 85.9 | |
| 39 | 85.9 | |
| 40 | 85.6 | |
| 41 | 85.3 | |
| 42 | 85.2 | |
| 43 | 85.2 | |
| 44 | 85.2 | |
| 45 | 84.9 | |
| 46 | 84.8 | |
| 47 | 84.8 | |
| 48 | 84.6 | |
| 49 | 84.5 | |
| 50 | 84.5 |