Long-horizon Task Completion on Minecraft Long-horizon Task Suite (held-out)
69.7Success RatePEAM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PEAMMemory type=parametric (MoE)2026.05 | 69.7 | 53 | 15.2 | 3.2 | 4.6 | |
| B2 Optimus-1-rep.Memory type=multimodal retrieval2026.05 | 60.6 | 43.7 | 6.1 | 6.1 | 28.4 | |
| B1 VOYAGERMemory type=retrieval2026.05 | 54.5 | 38 | — | 5.5 | 31.2 | |
| B5 EWCMemory type=parametric (regularized)2026.05 | 51.5 | 34.8 | 3 | 3.8 | — | |
| B4 Single shared LoRAMemory type=parametric (shared)2026.05 | 48.5 | 32 | 6 | 3.5 | — | |
| B3 Naive full-FTMemory type=parametric (full FT)2026.05 | 42.4 | 26.9 | 12.1 | 4.8 | — | |
| B8 MrSteveMemory type=spatial-temporal2026.05 | 33.3 | 18 | 21.2 | — | — | |
| B7 ReflexionMemory type=textual reflection2026.05 | 27.3 | 13.3 | 27.2 | — | — | |
| B6 No-memory ReActMemory type=none2026.05 | 6.1 | 0.7 | 48.4 | — | — |