Meta 纯推理模型获五项奥赛金牌，两项物理满分
Meta 纯推理 AI 在五科奥赛夺金，证明关掉工具也能达到顶尖推理水平，为下一代 Muse 模型热身。
Meta 宣布其 AI 模型在五项 STEM 学科奥赛中全部达到金牌水平，其中两项物理竞赛理论考试满分。模型被禁用搜索、编码、计算器等工具，纯靠推理答题。
正文摘录
Title: Is Meta back? Pure reasoning wins five STEM Olympiad gold medals, two with perfect scores Source: 机器之心 Body: .png) Editor: Panda Early this morning, [OpenAI made GPT-5.6 Luna free](https://mp.weixin.qq.com/s?biz=MzA3MzI4MjgzMw==&mid=2651048906&idx=2&sn=d0e7f21c94fc91c70f6b367f2af46232&scene=21wechatredirect). At almost the same time, another company that once created the Llama series but later fell behind in the large model race also made an announcement — Meta declared on 𝕏 that it had entered its AI models into five STEM subject Olympiads, achieving gold medal or gold-level results in all of them, including perfect scores on the theoretical exams of two …