Agentic Search on BrowseComp (Accuracy, Cost)
59.3AccuracySELFCOMPACT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SELFCOMPACTBackbone=MiniMax-M2.52026.06 | 59.3 | 0.09 | |
| SELFCOMPACTBackbone=Mimo-V2-Flash2026.06 | 57.9 | 0.11 | |
| Fixed-intervalBackbone=MiniMax-M2.52026.06 | 55.3 | 0.07 | |
| Keep-last-NBackbone=MiniMax-M2.52026.06 | 54.5 | 0.06 | |
| Delete-allBackbone=MiniMax-M2.52026.06 | 52.7 | 0.05 | |
| Fixed-intervalBackbone=Mimo-V2-Flash2026.06 | 51.6 | 0.14 | |
| Keep-last-NBackbone=Mimo-V2-Flash2026.06 | 49.8 | 0.06 | |
| No CompactionBackbone=MiniMax-M2.52026.06 | 47.2 | 0.19 | |
| Delete-allBackbone=Mimo-V2-Flash2026.06 | 45.5 | 0.08 | |
| No CompactionBackbone=Mimo-V2-Flash2026.06 | 42.7 | 0.27 | |
| SELFCOMPACTBackbone=GLM-4.7-Flash2026.06 | 41.2 | 0.1 | |
| Fixed-intervalBackbone=GLM-4.7-Flash2026.06 | 35.4 | 0.06 | |
| Keep-last-NBackbone=GLM-4.7-Flash2026.06 | 33.5 | 0.05 | |
| Delete-allBackbone=GLM-4.7-Flash2026.06 | 31.6 | 0.03 | |
| No CompactionBackbone=GLM-4.7-Flash2026.06 | 30.1 | 0.14 |