Agentic Task Success on MiniHack
20Success RateALMA
Evaluation Results
| Method | Links | |
|---|---|---|
| ALMAFoundation Model=GPT-5-mini2026.02 | 20 | |
| Trajectory RetrievalFoundation Model=GPT-5-mini2026.02 | 16.7 | |
| Reasoning BankFoundation Model=GPT-5-mini2026.02 | 15.8 | |
| No MemoryFoundation Model=GPT-5-mini2026.02 | 15 | |
| G-MemoryFoundation Model=GPT-5-mini2026.02 | 14.2 | |
| ALMAFoundation Model=GPT-5-nano2026.02 | 11.7 | |
| Dynamic CheatsheetFoundation Model=GPT-5-mini2026.02 | 11.7 | |
| Reasoning BankFoundation Model=GPT-5-nano2026.02 | 9.8 | |
| Dynamic CheatsheetFoundation Model=GPT-5-nano2026.02 | 9.2 | |
| Trajectory RetrievalFoundation Model=GPT-5-nano2026.02 | 7.5 | |
| G-MemoryFoundation Model=GPT-5-nano2026.02 | 6.8 | |
| No MemoryFoundation Model=GPT-5-nano2026.02 | 6.7 |