DeepSeek API's peak and off-peak pricing policy will be officially implemented starting at 0:00 on August 17. The cost of API calls during off-peak hours is only half of that during peak hours, aiming to guide developers to schedule large model tasks during off-peak times through market-based pricing mechanisms, thus alleviating daytime computing power congestion and improving platform stability.
The official peak hours are set as 9:00 to 12:00 and 14:00 to 18:00 Beijing Time daily, with the remaining time considered off-peak. The billing unit is unified as RMB per million tokens. The latest pricing for two main models takes effect simultaneously: During peak hours, the input (cache hit/miss) and output prices for DeepSeek V4Flash are 0.10 RMB, 3.0 RMB, and 9.0 RMB respectively, while during off-peak hours they are half of these values; during peak hours, the three prices for DeepSeek V4Pro are 0.30 RMB, 9.0 RMB, and 27.0 RMB respectively, and during off-peak hours they are 0.15 RMB, 4.5 RMB, and 13.5 RMB respectively.
Compared to the initial pricing when DeepSeek V4Pro was officially launched on August 12—input (cache miss) 3 RMB, output 6 RMB, and cache hit input as low as 0.025 RMB—the new policy has led to a 1100% increase in the price of cache hit input during peak hours, a 350% increase in output price, and a 200% increase in cache miss input price. Even when calling during off-peak hours, the three types of prices are 6 times, 1.5 times, and 2.25 times the original prices, respectively.
This implementation of the peak and off-peak pricing mechanism marks DeepSeek's formal departure from its previous "extreme cost-effectiveness" pricing strategy, offering a new reference model for fine-grained operation of computing power on domestic AI large model platforms. Notably, even with the adjusted peak prices, the output unit price of V4-Pro is still about one-thirteenth of that of Claude's flagship model, and during off-peak hours, it is about one-twenty-fifth.
