Long-Context Memory Retrieval and Reasoning on LongMemEval 50K context input token limit
89.7KU L-J ScoreHORMA
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HORMAPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=External Dynamic Memory, Context Token Limit=50K2026.06 | 89.7 | 58.9 | 13.3 | 88.6 | 33.1 | 55.9 | |
| AconPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=Internal Dynamic Memory, Context Token Limit=50K2026.06 | 64.1 | 26.8 | 40 | 61.4 | 21.8 | 40.6 | |
| A-MEMPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=External Dynamic Memory, Context Token Limit=50K2026.06 | 64.1 | 57.1 | 33.3 | 78.6 | 32.3 | 51.8 | |
| TruncationPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=Internal Static Memory, Context Token Limit=50K2026.06 | 62.8 | 37.5 | 6.7 | 42.9 | 17.3 | 34.1 | |
| ReSumPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=Internal Dynamic Memory, Context Token Limit=50K2026.06 | 60.3 | 19.6 | 33.3 | 52.9 | 15.8 | 34.3 | |
| Embedding RetrievalPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=External Dynamic Memory, Context Token Limit=50K2026.06 | 44.9 | 21.4 | 10 | 51.4 | 18.8 | 30.2 | |
| Mem0Primary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=External Dynamic Memory, Context Token Limit=50K2026.06 | 42.3 | 53.6 | 10 | 85.7 | 25.6 | 43.6 | |
| No limitPrimary Agent Backbone=Claude Sonnet 4.5, Context Token Limit=No limit2026.06 | 37.2 | 21.4 | 3.3 | 20 | 14.3 | 20.4 | |
| EmbeddingPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=Internal Static Memory, Context Token Limit=50K2026.06 | 37.2 | 12.5 | 3.3 | 35.7 | 17.3 | 23.2 | |
| Slide WindowPrimary Agent Backbone=Claude Sonnet 4.5, Memory Architecture Category=Internal Static Memory, Context Token Limit=50K2026.06 | 19.2 | 8.9 | 0 | 25.7 | 15.8 | 16.1 |