Fact Consolidation (Single-Hop) on MemoryAgentBench FC-SH Average
0.948AccuracyAblation C (gpt-4o backbone)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ablation C (gpt-4o backbone)Pipeline=Ablation C (gpt-4o backbone), Backbone=gpt-4o2026.05 | 0.948 | 0.921 | |
| Ablation A (chunk-4096)Pipeline=Ablation A (chunk-4096), Backbone=gpt-4o-mini2026.05 | 0.808 | 0.766 | |
| HeadlinePipeline=Headline, Backbone=gpt-4o-mini2026.05 | 0.78 | 0.737 | |
| LLM-judgment baselinePipeline=LLM-judgment baseline, Backbone=gpt-4o-mini2026.05 | 0.672 | 0.625 |