Memory-augmented generation on LongMemEvalS (test)
98.4User LLM (Single-Session)Full-Context
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Full-ContextBackbone=DeepSeek-V4-Flash2026.07 | 98.4 | 18.4 | 90 | 15.1 | 98.2 | 28.5 | 67.8 | 5.8 | 73.2 | 13.5 | 81.9 | 12.1 | 80.6 | 13.9 | |
| TRACEBackbone=DeepSeek-V4-Flash2026.07 | 93.8 | 79 | 83.3 | 41.3 | 96.4 | 79.8 | 53.4 | 44.5 | 65.6 | 27 | 79.6 | 56.6 | 73.3 | 50.3 | |
| NemoriBackbone=GPT-4o-mini2026.07 | 93.3 | 76.8 | 32.8 | 11.3 | 60.4 | 50.5 | 61.3 | 25.4 | 47.6 | 34.4 | 75.6 | 44.6 | 62.2 | 39.9 | |
| A-MemBackbone=DeepSeek-V4-Flash2026.07 | 92.2 | 76.7 | 35.6 | 10.1 | 100 | 82.8 | 40.8 | 25.4 | 44.9 | 37.2 | 61.6 | 42 | 58.8 | 44 | |
| TRACEBackbone=GPT-4o-mini2026.07 | 79.7 | 64.1 | 90 | 45.1 | 96.4 | 75.5 | 40.2 | 34.3 | 60.6 | 32.9 | 77.3 | 47.9 | 66.7 | 45.7 | |
| Full-ContextBackbone=GPT-4o-mini2026.07 | 78.1 | 17.6 | 48.9 | 11.2 | 79.8 | 16.6 | 43.3 | 3.5 | 40.2 | 12.7 | 73.6 | 12.2 | 56.5 | 11.3 | |
| A-MemBackbone=GPT-4o-mini2026.07 | 78.1 | 60 | 36.7 | 14 | 83.9 | 68.1 | 33.9 | 23.5 | 39.9 | 35.5 | 68.1 | 38.7 | 52.9 | 38.7 | |
| NemoriBackbone=DeepSeek-V4-Flash2026.07 | 66.7 | 60.4 | 37.4 | 11.4 | 31.9 | 27.6 | 42.3 | 38.6 | 74.8 | 46.4 | 64.7 | 43.7 | 56.3 | 41.4 |