AI 资讯 VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO 2026-06-23 原文 →