Overview of RL training systems and pass@k results
A December 21, 2025 publication covers on-policy rollout divergence, rollout-system design, and the effects of RL on pass@k performance.
Published on December 21, 2025, the overview covers divergence between on-policy rollouts, rollout-system design, and studies of reinforcement learning (RL) effects on pass@k results. Its description gives no methods, figures, or experimental conditions.
Reported results vary: gains are associated with more flexible reward attribution and DPO training of OLMo 3. The text also notes practical rollout challenges and emphasizes that RL’s effect on sampling performance depends on the training approach. Consult the original publication to verify the studies and their context; do not generalize findings beyond what each evaluation supports.