Multi-step web search on BrowseComp Chinese version (test)
34.6AccuracyLarge-model only
Evaluation Results
| Method | Links | |
|---|---|---|
| Large-model onlySpeedup=1.00x, Backbone=openPangu-DeepDiverV2-38B2025.12 | 34.6 | |
| AgentCollabSpeedup=1.32x, Small-model Backbone=openPangu-DeepDiverV2-7B, Large-model Backbone=openPangu-DeepDiverV2-38B2025.12 | 33.8 | |
| Small-model w/ large-model as InfoSeekerSpeedup=1.24x, Small-model Backbone=openPangu-DeepDiverV2-7B, Large-model Backbone=openPangu-DeepDiverV2-38B2025.12 | 27.3 | |
| Small-model w/ large-model as PlannerSpeedup=1.39x, Small-model Backbone=openPangu-DeepDiverV2-7B, Large-model Backbone=openPangu-DeepDiverV2-38B2025.12 | 24.6 | |
| Small-model onlySpeedup=1.54x, Backbone=openPangu-DeepDiverV2-7B2025.12 | 18.3 |