Long-term dialogue memory on PerLTQA (test)
93.14AccuracyMemBuilder
Evaluation results
| Method | Links | |
|---|---|---|
| MemBuilderMemory construction model=Qwen3-4B, Training stage=SFT + RL, Answer model=Claude 4.5 Sonnet2026.01 | 93.14 | |
| MemBuilderMemory construction model=Claude 4.5 Sonnet, Answer model=Claude 4.5 Sonnet2026.01 | 92.59 | |
| MemBuilderMemory construction model=GPT-4.1, Answer model=Claude 4.5 Sonnet2026.01 | 91.74 | |
| MemBuilderMemory construction model=Qwen3-4B, Training stage=SFT, Answer model=Claude 4.5 Sonnet2026.01 | 91.67 | |
| MemBuilderMemory construction model=QwQ-32B, Answer model=Claude 4.5 Sonnet2026.01 | 88.96 | |
| MIRIXMemory construction method=Prompting-based, Answer model=Claude 4.5 Sonnet2026.01 | 83.11 | |
| MemBuilderMemory construction model=Qwen3-4B, Training stage=RL, Answer model=Claude 4.5 Sonnet2026.01 | 82.19 | |
| RAG-SessionMemory construction method=Retrieval-based, Answer model=Claude 4.5 Sonnet2026.01 | 79.21 | |
| RAG-UtteranceMemory construction method=Retrieval-based, Answer model=Claude 4.5 Sonnet2026.01 | 77.23 | |
| MemBuilderMemory construction model=Qwen3-4B, Training stage=Base, Answer model=Claude 4.5 Sonnet2026.01 | 76.85 | |
| Mem0Memory construction method=Prompting-based, Answer model=Claude 4.5 Sonnet2026.01 | 62.04 |