DeepSeek Eliminates Weekend Peak API Pricing
DeepSeek changed its API billing rules so all Saturday and Sunday usage is charged at off-peak rates, reducing costs for schedulable weekend workloads.
Why it matters: Time-based API pricing gives developers and FinOps teams a new way to reduce inference costs by scheduling flexible workloads around lower-priced windows.
DeepSeek has changed its API billing system so that all usage on Saturdays and Sundays is charged at off-peak rates.
The new rule took effect at midnight Beijing time on August 23 and removes the distinction between peak and off-peak pricing during weekends. Weekday peak periods remain in place.
DeepSeek introduced variable API pricing earlier in August, with peak rates set at twice the corresponding off-peak rates.
Under the weekend policy, workloads that would otherwise fall into a peak period can therefore cost 50% less when run during Saturday or Sunday in Beijing time.
That creates an optimization opportunity for AI developers. Batch inference, software testing, model evaluation, indexing and other workloads that do not need to execute immediately can increasingly be scheduled around model-provider pricing windows.
The change illustrates how competition among AI providers is becoming more dynamic. Developers already compare models based on capability, latency and token cost. Variable pricing adds another factor: the same model and workload can have a different cost depending on when it runs.
As AI workloads scale, time-aware model routing could become another practical tool for controlling enterprise inference spending.