Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation

Benchmark Model Rank Results
code-generation-on-livecodebench-v6Road MOPD (Qwen3-4B)–Pass@1: 0.2957
mathematical-reasoning-on-aime-2025Road MOPD (Qwen3-4B)–Accuracy: 57.4
mathematical-reasoning-on-aime24Road MOPD (Qwen3-4B)–Acc: 61.88
mathematical-reasoning-on-hmmt-feb-2025Road MOPD (Qwen3-4B)–Accuracy: 33.54
medical-question-answering-on-medxpertqa-textRoad MOPD (Qwen3-4B)–Accuracy: 15.96
multiple-choice-question-answering-mcqa-on-medmcqaRoad MOPD (Qwen3-4B)–Test Set (Acc-%): 0.6311