Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data

Open paper
Benchmark Model Rank Results
mathematical-reasoning-on-aime-2025Mixed SFT + RLVR–Accuracy: 85.73
mathematical-reasoning-on-aime-2026Mixed SFT + RLVR–Accuracy: 70.42
mathematical-reasoning-on-aime24Mixed SFT + RLVR–Acc: 87.5
mathematical-reasoning-on-hmmt-feb-2025Mixed SFT + RLVR–Accuracy: 55.0
mathematical-reasoning-on-hmmt-feb-2026Mixed SFT + RLVR–Accuracy: 51.52
multi-task-language-understanding-on-mmlu-proMixed SFT + RLVR–Accuracy: 75.84
multiple-choice-question-answering-mcqa-on-gpqa-diamondMixed SFT + RLVR–Accuracy: 60.98
question-answering-on-humanity-s-last-exam-hleMixed SFT + RLVR–Accuracy: 9.24

Rank counts only results with a code link.