ResearchBenchmarksOffline Multi-Agent Sequential Decision Making on SMAC unseen tasks v2Follow67Win Rate (Protoss 5v5)DLM-GRPO-2.6815.4133.551.59Apr 26, 2026Evaluation ResultsMethodMethodLinksWin Rate (Protoss 5v5)Win Rate (Terran 5v5)Win Rate (Zerg 5v5)DLM-GRPOEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.04677958DLM-SFTEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.04596442MADTEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.04000GATOEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.04000