- DeepSeek introduces peak-and-off-peak API pricing, doubling rates during busy hours
- The move marks a shift from ultra-low pricing toward market-based management of scarce compute
DeepSeek’s new API pricing took effect on August 17, introducing a peak-and-off-peak pricing model for the first time among major Chinese foundation model providers.
Under the new system, peak hours run from 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time. Prices during these periods are twice the base rate, while off-peak rates are half the peak price.
The biggest increase applies to V4-Pro‘s cached-input pricing, which rises from 0.025 yuan to 0.30 yuan per million tokens during peak hours — an increase of 1,100%.
Uncached input rises 200% to 9 yuan, while output rises 350% to 27 yuan per million tokens.
V4-Flash also sees significant increases, with peak-hour prices rising 400% for cached input, 200% for uncached input and 350% for output.
The changes apply only to API usage. DeepSeek’s website and app remain free for ordinary users.
The impact will therefore fall primarily on developers and businesses that embed DeepSeek models into their own products or use them as production tools.
A shift in strategy
The new pricing also signals a broader shift in DeepSeek’s strategy. Rather than charging a uniform rate around the clock, the company is using price incentives to steer non-urgent workloads toward nights, weekends and holidays, when computing capacity is less constrained.
In effect, DeepSeek is bringing the logic of peak-and-off-peak electricity pricing to AI inference.
The move marks a notable departure from DeepSeek’s earlier reputation for ultra-low API prices — and highlights the growing pressure on leading AI providers to manage inference costs, capacity and demand more commercially.
