Starting at 0:00 on August 17, the new DeepSeek-4 API pricing will officially take effect. Time-sharing pricing was implemented in this round of adjustments. During peak periods, the output price rose to 27 yuan/million tokens, which is 4.5 times the previous price; the cache missed price increased to 9 yuan/million tokens, which is 3 times the previous price; and the cache hit price was drastically raised to 0.30 yuan/million tokens, which is 12 times the previous price. Prices during free time are half as high as during peak hours. After the price increase, DeepSeek said goodbye to the old “ultimate cost performance” route. Some industry insiders told the reporter that the previous SmartSeek model was relatively expensive because it could solve coding problems very well; DeepSeek focuses on cost performance, especially for many smart devices that operate 24 hours a day, which consumes an extremely large number of tokens, and developers often choose the DeepSeek model. However, DeepSeek's price increase is not an exception. Over the past few months, major domestic models such as Smart Spectrum, Kimi, and MiniMax have also successively raised the prices of some APIs. “This round of price adjustments is not a uniform price increase for the entire industry. Instead, independent model manufacturers are leading the way, and big manufacturers with their own computing power keep their hands on their hands or adjust prices in disguise.” Liang Wei, partner of Grant Thornton Consulting and Transaction Support Services, told the 21st Century Economic Report reporter that since this year, independent vendors such as Smart Spectrum, Dark Side of the Moon, and DeepSeek have successively raised API pricing, while major companies such as Ali, Byte, and Tencent have remained on hold.

Starting at 0:00 on August 17, the new DeepSeek-4 API pricing will officially take effect. Time-sharing pricing was implemented in this round of adjustments. During peak periods, the output price rose to 27 yuan/million tokens, which is 4.5 times the previous price; the cache missed price increased to 9 yuan/million tokens, which is 3 times the previous price; and the cache hit price was drastically raised to 0.30 yuan/million tokens, which is 12 times the previous price. Prices during free time are half as high as during peak hours. After the price increase, DeepSeek said goodbye to the old “ultimate cost performance” route. Some industry insiders told the reporter that the previous SmartSeek model was relatively expensive because it could solve coding problems very well; DeepSeek focuses on cost performance, especially for many smart devices that operate 24 hours a day, which consumes an extremely large number of tokens, and developers often choose the DeepSeek model. However, DeepSeek's price increase is not an exception. Over the past few months, major domestic models such as Smart Spectrum, Kimi, and MiniMax have also successively raised the prices of some APIs. “This round of price adjustments is not a uniform price increase for the entire industry. Instead, independent model manufacturers are leading the way, and big manufacturers with their own computing power keep their hands on their hands or adjust prices in disguise.” Liang Wei, partner of Grant Thornton Consulting and Transaction Support Services, told the 21st Century Economic Report reporter that since this year, independent vendors such as Smart Spectrum, Dark Side of the Moon, and DeepSeek have successively raised API pricing, while major companies such as Ali, Byte, and Tencent have remained on hold.