DeepSeek、モデルAPIにピーク・オフピーク料金を導入。要点は明快だ。DeepSeek’s V4 API pricing is framed like electricity: higher during busy windows, cheaper when demand is lower.
重要なのは次の層だ。The mechanism pushes developers to schedule batch and non-urgent inference when capacity is cheaper.
次に問われるのは実行だ。It also makes visible a cost truth most model vendors hide: inference capacity is a perishable resource.
参考資料: CocoLoopおよび原文で引用された公開報道