Fact Consolidation (Multi-Hop) on MemoryAgentBench (MAB) FC-MH 262K
27AccuracyAblation B (CAR, multi-hop)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ablation B (CAR, multi-hop)Pipeline=Ablation B (CAR, multi-hop), Backbone=gpt-4o-mini2026.05 | 27 | |
| GPT-4o (long-context)system=long-context2026.05 | 5 | |
| HippoRAG-v2 (best published)2026.05 | 5 | |
| BM25 (MAB chunk-512 default)2026.05 | 3 | |
| Zep / Graphiti (temporal KG)2026.05 | 3 |