Static Multi-Session QA on LONGMEMEVAL Oracle (clean)
98.21SS-AOBLIVION
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| OBLIVIONModel=GPT-4o-mini, Method Type=ours2026.03 | 98.21 | 94.29 | 73.33 | 81.2 | 80.77 | 83.46 | 85 | |
| OBLIVIONModel=GPT-4.1-mini, Method Type=ours2026.03 | 98.21 | 95.71 | 100 | 90.23 | 96.15 | 93.98 | 94.4 | |
| LME-RFTModel=GPT-4o-mini, Method Type=direct2026.03 | 91.07 | 91.43 | 83.33 | 80.45 | 80.77 | 78.95 | 83 | |
| LME-RFTModel=GPT-4.1-mini, Method Type=direct2026.03 | 91.07 | 94.29 | 100 | 90.98 | 92.31 | 89.47 | 91.8 | |
| EVERMEMOSModel=GPT-4.1-mini, Method Type=memory2026.03 | 83.93 | 96.67 | 96.67 | 78.7 | 92.31 | 74.19 | 83.8 | |
| EVERMEMOSModel=GPT-4o-mini, Method Type=memory2026.03 | 67.86 | 93.81 | 96.67 | 68.17 | 78.21 | 59.9 | 72.8 |