Long-term memory assessment on LONGMEMEVAL
86.46LLM-Judge Multi-sessionMRAgent*
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MRAgent*Retrieval Model=Claude, Memory Construction Model=Gemini2026.06 | 86.46 | 92.85 | 85.71 | 78.57 | 86.76 | |
| MRAgentBackbone=Gemini2026.06 | 68.42 | 92.85 | 68.42 | 66.67 | 72.95 | |
| MRAgent*Retrieval Backbone=Claude, Memory Construction Backbone=Gemini2026.06 | 66.31 | 82.41 | 60.1 | 15.58 | 86.76 | |
| MemoryOSBackbone=Gemini2026.06 | 56.39 | 87.14 | 38.35 | 46.67 | 54.92 | |
| RAGBackbone=Gemini2026.06 | 54.89 | 85.71 | 42.86 | 33.33 | 54.65 | |
| LangMemBackbone=Gemini2026.06 | 52.63 | 78.57 | 45.71 | 36.67 | 53.77 | |
| Mem0Backbone=Gemini2026.06 | 50.38 | 78.57 | 45.11 | 40 | 53.01 | |
| MRAgentBackbone=Gemini2026.06 | 49.92 | 80.99 | 50.16 | 23.96 | 72.95 | |
| MemoryOSBackbone=Gemini2026.06 | 45.02 | 79.5 | 35.67 | 9.53 | 54.92 | |
| RAGBackbone=Gemini2026.06 | 45 | 77.94 | 43.88 | 5.42 | 54.65 | |
| A-MemBackbone=Gemini2026.06 | 42.85 | 90 | 45.11 | 46.43 | 52.98 | |
| LangMemBackbone=Gemini2026.06 | 41.14 | 72.43 | 36.8 | 6.32 | 53.77 | |
| Mem0Backbone=Gemini2026.06 | 37.53 | 72.43 | 35.64 | 5.89 | 53.01 | |
| A-MemBackbone=Gemini2026.06 | 28.26 | 75.3 | 34.18 | 9.09 | 52.98 |