Direct On-Policy Distillation Cuts the Cost of RLVR on New Strong Models

arXiv·medium signal

Reinforcement learning with verifiable rewards is expensive to re-run on every new strong model because the target must generate many rollouts; this work distills on-policy from a weaker model directly to achieve weak-to-strong generalization more cheaply. It is an efficiency win for teams repeatedly doing reasoning post-training as models scale. Practical when rollout compute dominates your training bill.