List your company

Agentic Task Success on MiniHack

20Success RateALMA
6.1689.75913.3516.941Feb 8, 2026

Evaluation results

MethodLinks
ALMAFoundation Model=GPT-5-mini2026.0220
Trajectory RetrievalFoundation Model=GPT-5-mini2026.0216.7
Reasoning BankFoundation Model=GPT-5-mini2026.0215.8
No MemoryFoundation Model=GPT-5-mini2026.0215
G-MemoryFoundation Model=GPT-5-mini2026.0214.2
ALMAFoundation Model=GPT-5-nano2026.0211.7
Dynamic CheatsheetFoundation Model=GPT-5-mini2026.0211.7
Reasoning BankFoundation Model=GPT-5-nano2026.029.8
Dynamic CheatsheetFoundation Model=GPT-5-nano2026.029.2
Trajectory RetrievalFoundation Model=GPT-5-nano2026.027.5
G-MemoryFoundation Model=GPT-5-nano2026.026.8
No MemoryFoundation Model=GPT-5-nano2026.026.7