DeepSeek 給大模型 API 搞峰谷電價。這篇文章的核心事實很清楚：DeepSeek’s V4 API pricing is framed like electricity: higher during busy windows, cheaper when demand is lower.
更值得看的，是第二層含義：The mechanism pushes developers to schedule batch and non-urgent inference when capacity is cheaper.
接下來要看的，是執行能否跟上。It also makes visible a cost truth most model vendors hide: inference capacity is a perishable resource.
參考來源: CocoLoop及原文引用的公開報導