Chinese company DeepSeek, which became known for offering low-cost AI models and competing with pricier Western products, intends to significantly raise the API prices for its latest models.
The company, concurrent with the launch of DeepSeek V4 Pro, announced that from 16 August it will implement a new pricing structure for its API services. Under this change, usage costs will differ between peak and off-peak hours, and prices for some services will increase up to four times.
DeepSeek cited more rational resource allocation as the reason for this change; an action that suggests the era of offering AI services at very low prices may be gradually coming to an end.
Fourfold price increase for DeepSeek V4 Pro
Under the new prices, using DeepSeek V4 Pro to generate one million output tokens during peak hours will cost $3.96. This figure represents more than a fourfold increase compared to the current price of $0.87. During off-peak hours, using the same model will cost $1.98 per one million output tokens; that is half the price set for peak hours.
The DeepSeek V4 Flash model will also see a price increase. The cost to generate one million output tokens with this model during peak hours will rise from the current $0.28 to $1.32. In off-peak hours users will have to pay $0.66 per one million output tokens.
End of promotional pricing
DeepSeek's current prices were initially offered as part of a promotional scheme and were supposed to end on 31 May. However, DeepSeek announced in the same month that it intended to keep the discounted prices permanently.
Despite that decision, the company ultimately took a different path and, with the introduction of the V4 models, moved forward with raising API service prices. The new time-based pricing structure may also help DeepSeek better manage high demand during busy hours and allocate its compute capacity according to user consumption.
DeepSeek remains among the low-cost models
Despite the fourfold price increase, DeepSeek's new models still remain relatively inexpensive compared to many competing models. For example, the Kimi K3 model from Chinese company Moonshot costs $15 per one million output tokens. In contrast, OpenAI's GPT-5.6 Sol is priced at $30 for the same output.
Of course, among the cheaper models, GPT-5.6 Luna at $1.20 per one million output tokens still costs less than DeepSeek V4 Flash even during peak hours.
Thus, although DeepSeek's price increase may erode a significant part of the company's cost advantage, the V4 models still offer more competitive pricing compared with many of the market's advanced models.