- The move comes just a week after DeepSeek introduced peak and off-peak pricing, with V4-Pro rates rising by as much as 12 times for some token categories
- The weekend discount gives developers a cheaper window for batch workloads while allowing DeepSeek to shift demand away from busy weekday hours
DeepSeek has begun charging its lower off-peak API rates throughout Saturdays and Sundays, giving developers a 50% discount on weekend usage just one week after the Chinese AI startup introduced a new peak-and-off-peak pricing system.
Under the revised rules, effective from midnight Beijing time on August 23, DeepSeek no longer distinguishes between peak and off-peak hours on weekends. All Saturday and Sunday API calls are billed at the lower off-peak rate.
For DeepSeek-V4-Pro, output during weekday peak hours costs 27 yuan ($4) per million tokens, compared with 13.5 yuan during off-peak periods.
That means developers now pay 13.5 yuan throughout the weekend, including during what would normally be peak hours. V4-Flash output is similarly priced at 4.5 yuan per million tokens on weekends.
The weekday schedule remains unchanged. Peak hours run from 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time, while all other weekday hours are off-peak. Off-peak rates are set at half the peak rates.
Major overhaul of API pricing
The weekend change comes only days after DeepSeek made a major overhaul of its API pricing. From August 17, the company introduced time-based pricing for its V4 models, with peak rates set at twice off-peak prices.
Some V4-Pro rates rose as much as 1,100% from previous levels, according to Reuters.
That abrupt sequence — a steep price increase followed by a weekend discount — highlights the delicate balancing act facing DeepSeek as it expands commercial API usage while managing limited inference capacity.
The new pricing also gives developers a clear incentive to reschedule workloads. Batch inference, data processing, model evaluation and other tasks that do not require immediate responses can be pushed to weekends without paying the weekday peak premium.
For DeepSeek, the strategy could help smooth demand rather than simply cut prices. By charging more when computing demand is high and less when capacity is available, the company is effectively applying a version of peak-and-off-peak utility pricing to AI inference.
Why it matters globally
The shift also marks a broader change in China’s AI API market. DeepSeek had built much of its reputation on exceptionally low prices, but its latest moves suggest that leading model providers are increasingly treating inference capacity as a scarce resource that needs to be actively managed.
For international developers and enterprises, the experiment offers a glimpse of where AI pricing could be heading: away from a simple per-token rate and toward pricing that reflects not only what model is used, but when the computing power is consumed.
