VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO(arxiv.org)
357 points by timhigins | 185 comments
arXiv technical report introducing VibeThinker-3B, a 3B-parameter dense model trained with curriculum SFT, multi-domain RL, and self-distillation for verifiable reasoning.