List your company

Direct-Answer Question Answering on RefMem-Bench

32.9AccuracyREMIND
11.78817.26922.7528.231May 31, 2026

Evaluation results

MethodLinks
REMINDBackbone=Qwen3-VL-8B2026.0532.955.427.630.4
Qwen3-VL-30BBackbone=30B2026.0523.640.419.522.4
LightMemBackbone=GPT-4o-mini2026.0523.239.920.123.2
GAMBackbone=GPT-4o-mini2026.0523.236.819.923
A-MemBackbone=GPT-4o-mini2026.0522.939.616.418.5
Mem0Backbone=GPT-4o-mini2026.0521.8—17.519.9
Base ModelBackbone=Qwen3-VL-8B2026.0521.137.917.821.3
LangMemBackbone=GPT-4o-mini2026.0521.1—19.922.3
- w/ NaiveRAGBackbone=Qwen3-VL-8B2026.0519.838.71719.8
DS-Distill-Qwen-14BBackbone=14B2026.0519.139.115.418.8
MemoryOSBackbone=GPT-4o-mini2026.051935.618.621.2
MemGPTBackbone=GPT-4o-mini2026.0517.334.316.519.3
MEM1-7BBackbone=7B2026.0512.724.610.712.9
MemAgent-7BBackbone=7B2026.0512.629.712.114.2