ResearchBenchmarksOffline Multi-Agent Sequential Decision Making on SMAC unseen tasksFollow78Win Rate (3s vs 3z)DLM-GRPO43.6852.5961.570.41Apr 26, 2026Evaluation ResultsMethodMethodLinksWin Rate (3s vs 3z)Win Rate (3s vs 4z)Win Rate (3m)Win Rate (8m)Win Rate (25m)Win Rate (MMM)Win Rate (1o 10b vs 1r)Win Rate (1o 2r vs 4r)DLM-GRPOEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.04788293100991006469DLM-SFTEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.0471799010098995764GATOEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.0452678854796872MADTEvaluation Protocol=Ze...Evaluation Protocol=Zero-shot2026.044573926957851311