Multi-Hop Fact-based Reasoning on MAB FC-MH, 262K v3 (test)
27AccuracyChain-Aware Resolution (CAR)
Evaluation Results
| Method | Links | |
|---|---|---|
| Chain-Aware Resolution (CAR)Backbone=gpt-4o-mini, Pipeline architecture=multi-hop2026.05 | 27 | |
| GPT-4o (long-context)Backbone=gpt-4o, Pipeline architecture=long-context2026.05 | 5 | |
| HippoRAG-v2 (best published)Backbone=gpt-4o-mini2026.05 | 5 | |
| BM25Backbone=gpt-4o-mini, Pipeline architecture=MAB chunk-512 default2026.05 | 3 | |
| Zep / GraphitiBackbone=gpt-4o-mini, Pipeline architecture=temporal KG2026.05 | 3 |