Stratagem: Learning Transferable Reasoning via Trajectory-Modulated Game Self-Play

Benchmark Model Rank Results
mathematical-reasoning-on-olympiadbenchStratagem (All Games)#44Accuracy: 39.9
multi-task-language-understanding-on-mmlu-proStratagem (All Games)#101Accuracy: 57.83
mathematical-reasoning-on-aime24Stratagem (All Games)#119Acc: 20.0
mathematical-reasoning-on-aime-2025Stratagem (Kuhn Poker)#139Accuracy: 13.3