Claude 3.7 Sonnet 刚上线的时候我就直接问了几道数学题目，发现官网开 Thinking 似乎也不太行 
直到我今天晚上翻了翻官网的 Announcement，看到了一句话 

Third, in developing our reasoning models, we’ve optimized somewhat less for math and computer science competition problems, and instead shifted focus towards real-world tasks that better reflect how businesses actually use LLMs. 

我就在想，欸你说这 Anthropic 会不会说的是真的，于是我就掏出了前几天所有 AI 模型都没有成功解决的一个问题： 
Visualize ...