Single-Choice Question Answering on RefMem-Bench
66.2AccuracyREMIND
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| REMINDBackbone=Qwen3-VL-8B2026.05 | 66.2 | 52.3 | |
| Qwen3-VL-30BBackbone=30B2026.05 | 58.3 | 40.5 | |
| LangMemBackbone=GPT-4o-mini2026.05 | 57.8 | — | |
| A-MemBackbone=GPT-4o-mini2026.05 | 57.4 | 44.2 | |
| LightMemBackbone=GPT-4o-mini2026.05 | 56.9 | 44.1 | |
| GAMBackbone=GPT-4o-mini2026.05 | 56.4 | 41.4 | |
| DS-Distill-Qwen-14BBackbone=14B2026.05 | 54.9 | 45.7 | |
| MemAgent-7BBackbone=7B2026.05 | 53.5 | 43.9 | |
| MemoryOSBackbone=GPT-4o-mini2026.05 | 53.1 | 40.9 | |
| MEM1-7BBackbone=7B2026.05 | 49.2 | 36.9 | |
| Mem0Backbone=GPT-4o-mini2026.05 | 48.5 | — | |
| MemGPTBackbone=GPT-4o-mini2026.05 | 47.7 | 39.3 | |
| - w/ NaiveRAGBackbone=Qwen3-VL-8B2026.05 | 47.6 | 38.6 | |
| Base ModelBackbone=Qwen3-VL-8B2026.05 | 45 | 37.8 |