Long Multi-session Question Answering on LoCoMo 1.0 (test)
37.5Single-hop F1No limit
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| No limitBackbone=Claude Sonnet 4.5, Context Limit=Unconstrained2026.06 | 37.5 | 17.9 | 0.7 | 24 | |
| HORMABackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=External Dynamic Memory2026.06 | 32.3 | 16.6 | 7.6 | 21.4 | |
| Embedding RetrievalBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=External Dynamic Memory2026.06 | 26.5 | 15.7 | 0.4 | 15 | |
| Mem0Backbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=External Dynamic Memory2026.06 | 25.6 | 10.1 | 2.6 | 16.1 | |
| AconBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=Internal Dynamic Memory2026.06 | 23.2 | 10.1 | 9.7 | 10.4 | |
| A-MEMBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=External Dynamic Memory2026.06 | 23 | 12.2 | 1.9 | 13.2 | |
| TruncationBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=Internal Static Memory2026.06 | 20.5 | 10.8 | 0.5 | 12.7 | |
| EmbeddingBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=Internal Static Memory2026.06 | 20.2 | 12.2 | 0.5 | 13.4 | |
| Slide WindowBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=Internal Static Memory2026.06 | 15.5 | 13.3 | 0.7 | 11.2 | |
| ReSumBackbone=Claude Sonnet 4.5, Context Limit=10K tokens, Memory Architecture=Internal Dynamic Memory2026.06 | 11.7 | 10.6 | 0.4 | 8.6 |