Information Extraction on xbench-deepsearch Out-of-Distribution
47AccuracyClaude Sonnet 4
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude Sonnet 4Model Category=Closed-source Models2025.07 | 47 | 71 | 29 | |
| GPT-4.1Model Category=Closed-source Models2025.07 | 38 | 43 | 57 | |
| Deliberative Searcher-72BModel Category=70B Models, Backbone=Qwen2.5-VL-72B2025.07 | 35 | 77 | 9 | |
| GPT-4oModel Category=Closed-source Models2025.07 | 32 | 69 | 31 | |
| InternVL3-78BModel Category=70B Models, Backbone=InternVL3-78B2025.07 | 24 | 53 | 47 | |
| Qwen2.5-VL-72BModel Category=70B Models, Backbone=Qwen2.5-VL-72B2025.07 | 23 | 60 | 39 | |
| Deliberative Searcher-DeepSeek-70BModel Category=70B Models, Backbone=DeepSeek-R1-Distill-70B2025.07 | 18 | 80 | 8 | |
| R1-Searcher-7BModel Category=7B Models, Backbone=R1-7B2025.07 | 17 | 36 | 63 | |
| ReSearch-7BModel Category=7B Models2025.07 | 17 | 23 | 77 | |
| Deliberative Searcher-7BModel Category=7B Models, Backbone=Qwen2.5-VL-7B, Reward Function=ECE-based2025.07 | 16 | 86 | 1 | |
| Deliberative Searcher-7BModel Category=7B Models, Backbone=Qwen2.5-VL-7B2025.07 | 15 | 86 | 1 | |
| Search-R1-7BModel Category=7B Models, Backbone=R1-7B2025.07 | 14 | 48 | 52 | |
| DeepSeek-R1-Distill-70BModel Category=70B Models, Backbone=DeepSeek-R1-Distill-70B2025.07 | 14 | 40 | 60 | |
| Qwen2.5-VL-7BModel Category=7B Models, Backbone=Qwen2.5-VL-7B2025.07 | 12 | 58 | 42 |