On-policy Distillation with Verifiable Reward

Benchmark Model Rank Results
mathematical-reasoning-on-minerva-mathOPDVR (Qwen3-4B)#33Accuracy: 33.2
mathematical-reasoning-on-olympiadbenchOPDVR (Qwen3-4B)#34Accuracy: 47.0
mathematical-reasoning-on-math-500OPDVR (Qwen3-4B)#73Accuracy: 84.7
mathematical-reasoning-on-aime24OPDVR (Qwen3-4B)#97Acc: 36.9
mathematical-reasoning-on-aime-2025OPDVR (Qwen3-4B)#114Accuracy: 28.1