Long-context Conversation Reasoning on LoCoMo
82.62Multi-Hop AccuracyFullContext
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| FullContextBackbone=GPT-4.1-MINI2026.05 | 82.62 | 79.13 | 57.29 | 90.84 | 84.18 | 0 | 31.52 | |
| RecMemBackbone=GPT-4.1-MINI2026.05 | 78.37 | 75.39 | 57.29 | 86.92 | 81.1 | 193.2 | 2.75 | |
| RecMemBackbone=GPT-4O-MINI2026.05 | 72.7 | 57.32 | 58.33 | 79.79 | 72.47 | 202.4 | 2.73 | |
| FullContextBackbone=GPT-4O-MINI2026.05 | 69.86 | 52.96 | 55.21 | 90.01 | 76.43 | 0 | 31.5 | |
| MemoryOSBackbone=GPT-4.1-MINI2026.05 | 66.31 | 47.66 | 55.21 | 77.05 | 67.6 | 400.7 | 5.04 | |
| A-MemBackbone=GPT-4.1-MINI2026.05 | 59.93 | 72.9 | 42.71 | 73.25 | 68.83 | 1,459.93 | 5.56 | |
| MemoryOSBackbone=GPT-4O-MINI2026.05 | 58.87 | 44.24 | 46.88 | 74.55 | 63.64 | 429.7 | 4.88 | |
| Naive RAGBackbone=GPT-4.1-MINI2026.05 | 58.87 | 33.96 | 50 | 61.24 | 54.42 | 0 | 6.28 | |
| Mem0Backbone=GPT-4.1-MINI2026.05 | 58.16 | 63.86 | 44.79 | 66.23 | 62.92 | 1,520.8 | 2.11 | |
| Naive RAGBackbone=GPT-4O-MINI2026.05 | 54.61 | 24.3 | 51.04 | 57.55 | 49.68 | 0 | 6.23 | |
| Mem0Backbone=GPT-4O-MINI2026.05 | 52.84 | 52.02 | 37.5 | 65.52 | 58.64 | 1,233.5 | 1.99 | |
| A-MemBackbone=GPT-4O-MINI2026.05 | 52.13 | 61.99 | 30.21 | 66.83 | 60.84 | 1,143.3 | 3.04 |