DeepSeek removes weekend peak pricing after raising API costs

Signals Inbox·August 25, 2026·AI Infrastructure

DeepSeek has removed weekend peak pricing for its API after introducing much higher peak rates. The new schedule makes every weekend hour off-peak while weekdays still have expensive morning and afternoon windows. It is a small pricing change with a bigger message: inference capacity is scarce enough to be priced like electricity.

Q1What changed in DeepSeek's pricing?

DeepSeek's new schedule, summarized in Bloomberg reporting, makes weekends entirely off-peak. On weekdays, peak periods remain concentrated around 09:00-12:00 and 14:00-18:00 Beijing time, with lower rates outside those windows.

Q2Why did developers care so much?

Because the peak rates introduced earlier were a sharp break from DeepSeek's low-cost positioning. TechNode reported that V4 Flash output rose to RMB 9 per million tokens at peak versus RMB 4.5 off-peak, while V4 Pro reached RMB 27 versus RMB 13.5.

Q3Why use time-of-day pricing at all?

Inference infrastructure has a utilization problem. Servers are expensive whether they are busy or idle, and demand clusters around working hours. Charging more at peak and less off-peak gives developers a reason to move batch jobs, evaluations and non-urgent workloads into quieter periods.

Q4Does weekend off-peak pricing reverse the increase?

No. It softens it. Developers that can shift work to weekends get the lower rate all day, but weekday peak prices remain. The bigger pricing reset still stands, which means DeepSeek is prioritizing capacity management and monetization over being the cheapest option at every hour.

Q5What does this say about the AI API market?

Compute is becoming a dynamic commodity. As providers run larger models and face bursts of demand, flat token pricing can leave money on the table or overload capacity. Expect more cloud-like pricing behavior: tiers, reserved capacity, caching discounts and time-sensitive rates.