Agentic Search on BrowseComp-Plus (Accuracy, Cost)
71.2AccuracySELFCOMPACT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SELFCOMPACTBackbone=MiniMax-M2.52026.06 | 71.2 | 0.07 | |
| Keep-last-NBackbone=MiniMax-M2.52026.06 | 67.4 | 0.05 | |
| Fixed-intervalBackbone=MiniMax-M2.52026.06 | 65.9 | 0.04 | |
| Delete-allBackbone=MiniMax-M2.52026.06 | 65 | 0.03 | |
| Keep-last-NBackbone=Mimo-V2-Flash2026.06 | 63.5 | 0.15 | |
| SELFCOMPACTBackbone=Mimo-V2-Flash2026.06 | 62.9 | 0.16 | |
| No CompactionBackbone=MiniMax-M2.52026.06 | 62 | 0.19 | |
| Delete-allBackbone=Mimo-V2-Flash2026.06 | 61.2 | 0.11 | |
| Fixed-intervalBackbone=Mimo-V2-Flash2026.06 | 60.2 | 0.14 | |
| No CompactionBackbone=Mimo-V2-Flash2026.06 | 57.6 | 0.24 | |
| SELFCOMPACTBackbone=GLM-4.7-Flash2026.06 | 54.1 | 0.04 | |
| Keep-last-NBackbone=GLM-4.7-Flash2026.06 | 51.4 | 0.03 | |
| Fixed-intervalBackbone=GLM-4.7-Flash2026.06 | 50 | 0.04 | |
| Delete-allBackbone=GLM-4.7-Flash2026.06 | 47.3 | 0.02 | |
| No CompactionBackbone=GLM-4.7-Flash2026.06 | 45.6 | 0.12 |