Multimodal Deep Information Seeking on BrowseComp-VL
45.2L1 ScoreFlash-Searcher
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Flash-SearcherProtocol=Agent Workflow, Backbone=GPT-52026.06 | 45.2 | 51 | 47.9 | |
| Struct-SearcherProtocol=Agent Workflow, Backbone=GPT-52026.06 | 44.2 | 52.5 | 48.6 | |
| Struct-SearcherProtocol=Agent Workflow, Backbone=Gemini-2.5-Pro2026.06 | 39.7 | 36.5 | 38.3 | |
| Flash-SearcherProtocol=Agent Workflow, Backbone=Gemini-2.5-Pro2026.06 | 35.2 | 34 | 35.1 | |
| Gemini-2.5-ProProtocol=Direct Inference2026.06 | 34.7 | 28.5 | 31.3 | |
| Struct-SearcherProtocol=Agent Workflow, Backbone=GPT-4o2026.06 | 33.7 | 29 | 31.3 | |
| Flash-SearcherProtocol=Agent Workflow, Backbone=GPT-4o2026.06 | 32.2 | 24.5 | 28.1 | |
| GPT-5Protocol=Direct Inference2026.06 | 30.2 | 33.5 | 32.1 | |
| WebWatcher-32BProtocol=Agent Workflow2026.06 | 28.4 | 25 | 27 | |
| WebWatcher-7BProtocol=Agent Workflow2026.06 | 23.6 | 17 | 21.2 | |
| OmniSearchProtocol=Agent Workflow, Backbone=GPT-4o2026.06 | 19.7 | 10 | 16.3 | |
| Gemini-2.5-FlashProtocol=Direct Inference2026.06 | 11.6 | 6 | 9.6 | |
| Qwen2.5-VL-72BProtocol=Direct Inference2026.06 | 9.2 | 3 | 7.1 | |
| Claude-3.7-SonnetProtocol=Direct Inference2026.06 | 8.8 | 4 | 7.1 | |
| GPT-4oProtocol=Direct Inference2026.06 | 6.4 | 4 | 5.5 | |
| Qwen2.5-VL-32BProtocol=Direct Inference2026.06 | 3.2 | 1 | 2.4 | |
| Qwen2.5-VL-7BProtocol=Direct Inference2026.06 | 0.8 | 0 | 0.5 |