A million output tokens from DeepSeek's largest model cost $3.96 during Chinese office hours and $1.98 outside them. The split arrived with DeepSeek-V4-Pro-0813, and the rule is published plainly: off-peak rates are half of the peak rates, and peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC. In Beijing that is nine to noon and two to six, with lunch discounted.
Utilities price this way. A serving fleet costs the same whether it is busy or idle, demand for it follows the working day, and the cheapest revenue a provider can find sits in the trough. A discount that exists only at night costs the seller almost nothing and halves the bill for any buyer whose work can wait. Evaluation runs, document backfills and overnight extraction all can.
Our own basket ran into the clock. The snapshot is taken on a Monday morning in Europe, inside the second peak window. Our token table records $1.32 and $3.96 for the model, the expensive half of a split price. The seven snapshots before it all read a flat $0.435 and $0.87. Part of that gap is a real repricing with the new model version and part of it is the hour we asked, and the hour a European team asks at is the expensive one.
DeepSeek V4 Pro came back at $1.32 and $3.96 per million tokens against $0.435 and $0.87 in the seven previous snapshots, and the support-assistant floor rose to $383 a month from $113. The snapshot was taken at 09:37 UTC, inside a peak window.
