Lifelong Agent Learning on LifelongAgentBench (LAB) (test)
38.4DBMem-π
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Mem-πBase Agent=gpt-5.4-mini, Memory Backbone=Qwen-2.5-7B-Instruct2026.05 | 38.4 | — | 35 | 36.7 | |
| Mem-π (Stage 1)Base Agent=gpt-5.4-mini, Memory Backbone=Qwen-2.5-7B-Instruct, Training Stage=Stage 12026.05 | 35.3 | — | 32.9 | 34.1 | |
| MemRLBase Agent=gpt-5.4-mini2026.05 | 33 | — | 30.7 | 31.9 | |
| Memory-R1Base Agent=gpt-5.4-mini, Memory Backbone=Qwen-2.5-7B-Instruct2026.05 | 32.6 | — | 29.8 | 31.2 | |
| Mem0Base Agent=gpt-5.4-mini, k=12026.05 | 31.9 | — | 28.1 | 30 | |
| RAGBase Agent=gpt-5.4-mini, k=12026.05 | 29.9 | — | 27.2 | 28.5 | |
| Base AgentBase Agent=gpt-5.4-mini2026.05 | 28.5 | — | 25 | 26.8 |