@article{reuseyourflopsscalingrlonhardproblemsbyc, title = {Reuse your FLOPs: Scaling RL on Hard Problems by Conditioning on Very Off-Policy Prefixes}, author = {Amrith Setlur and Zijian Wang and Andrew Cohen and Paria Rashidinejad and Sang Michael Xie}, year = {2026}, eprint = {2601.18795}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2601.18795}, }