DeepSeek increases the cost of the software interface for its new AI models
8/14/2026, 03:47 PM • Евгения Слив

The Chinese company DeepSeek, previously famous for its noticeably more affordable artificial intelligence-based services compared to Western competitors, is starting to increase the cost of its products. Along with the official announcement of the new V4 Pro model, the developers warned customers about a significant increase in prices for the software interface. Since August 16, a new tariff system has been introduced, divided into peak and off-peak hours, which, according to the creators, will allow for a more rational allocation of computing resources. During peak hours, access to the V4 Pro model will cost three dollars and ninety-six cents for one million exit tokens, which is more than four times the previous rate of eighty-seven cents. During periods of lower load, the price will be half of this amount, that is, one dollar and ninety-eight cents.
The more affordable V4 Flash model will cost one dollar and thirty-two cents per million output tokens during peak hours instead of the previous twenty-eight cents, and the cost will decrease to sixty-six cents during off-peak hours. It is worth noting that the discounted tariffs in force up to this point were originally conceived as a temporary promotion, which was supposed to end in May. Then the company extended the reduced prices, but eventually decided to go for a significant increase. The new tariffs will take effect simultaneously with the launch of the updated version of the model.
Despite the fourfold increase in cost, the V4 family models remain significantly more affordable than many direct competitors on the market. For comparison, using the Kimi K3 neural network from the Chinese company Moonshot costs fifteen dollars per million output tokens. The most advanced model from OpenAI, called GPT-5.6 Sol, costs thirty dollars for a similar volume. At the same time, the budget version from the American company GPT-5.6 Luna is estimated at one dollar and twenty cents, which turns out to be even cheaper than using DeepSeek V4 Flash during peak load hours. Thus, DeepSeek retains its price advantage in the segment of advanced models, despite the revision of the tariff policy.
