Memory Management on Memalpha (val)
83.6SQuADFine-Mem
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Fine-MemAverage memory length per sample (k tokens)=10.2K2026.01 | 83.6 | 87.2 | 69.3 | 67.5 | 72.8 | 62.4 | 21.3 | 66.3 | |
| A-MemAverage memory length per sample (k tokens)=11.5K2026.01 | 82.7 | 82.6 | 22.7 | 62.8 | 68.6 | 44.3 | 18.1 | 54.5 | |
| TRUSTMEM2026.06 | 80.6 | 84.9 | 63.6 | 72.4 | 73.8 | 64 | 22.6 | 66.3 | |
| Mem-αAverage memory length per sample (k tokens)=7.90K2026.01 | 78.6 | 83.2 | 65.9 | 66.6 | 65.8 | 54.5 | 18.7 | 61.9 | |
| Mem-α2026.06 | 78.6 | 83.2 | 65.9 | 66.6 | 65.8 | 54.5 | 18.7 | 61.9 | |
| RAG-Top2Average memory length per sample (k tokens)=11.3K2026.01 | 76.2 | 84.9 | 62.3 | 61.2 | 50.8 | 57 | 4.2 | 56.7 | |
| RAG-Top22026.06 | 76.2 | 84.9 | 62.3 | 61.2 | 50.8 | 57 | 4.2 | 56.7 | |
| Long-ContextAverage memory length per sample (k tokens)=10.8K2026.01 | 74.2 | 85.2 | 60.5 | 62.3 | 70.8 | 53.3 | 5.2 | 58.8 | |
| Long-Context2026.06 | 74.2 | 85.2 | 60.5 | 62.3 | 70.8 | 53.3 | 5.2 | 58.8 | |
| MemAgent2026.06 | 52.6 | 66 | 41.5 | 53.5 | 57.4 | 61.6 | 8.6 | 48.7 | |
| LightMemAverage memory length per sample (k tokens)=2.12K2026.01 | 21.4 | 27.1 | 28.3 | 83.3 | 67.1 | 21.4 | 15.4 | 37.7 | |
| MEM12026.06 | 12 | 16.4 | 16 | 47.5 | 33.2 | 22.6 | 6.6 | 22 | |
| MemAgentAverage memory length per sample (k tokens)=0.84K2026.01 | 9.1 | 14 | 5.2 | 56.2 | 29 | 34.3 | 10.3 | 22.6 | |
| MEM1Average memory length per sample (k tokens)=0.17K2026.01 | 3.9 | 8.3 | 6.8 | 26.9 | 5.6 | 17.5 | 8.5 | 11.1 |