DeepSeek cuts off-peak Flash input prices by as much as 60 percent
DeepSeek will change the Flash series' off-peak tariff at noon Beijing time on 10 September. Per million tokens, cached input will cost 0.02 yuan, uncached input 1 yuan and output 4 yuan; peak rates are double. Yicai's comparison with current prices puts the reductions between about 11% and 60 percent.
Artificial Intelligence··Morning
All three billing components get cheaper
DeepSeek announced that it will change pricing for the Flash series from 12.00 Beijing time on 10 September. The tariff reported by 36Kr and Yicai sets off-peak prices per million tokens at 0.02 yuan for cached input, 1 yuan for uncached input and 4 yuan for output. Peak rates are twice those figures. The time zone and weekday definition matter: peak periods are 09.00-12.00 and 14.00-18.00 Beijing time on weekdays, while all other hours count as off-peak.[1], [2]
Cached input receives the largest reduction
Yicai compares the new prices with the current off-peak tariff. Cached input falls from 0.05 yuan to 0.02 yuan per million tokens, a 60 percent reduction. Uncached input declines from 1.5 yuan to 1 yuan, about 33 percent, while output moves from 4.5 yuan to 4 yuan, about 11 percent. The split means that a total bill cannot be calculated from one headline percentage. The effective reduction depends on how much of a workload hits the cache and how much output it produces.[2]
Scheduling becomes part of usage cost
The tariff makes the price of the same request depend not only on token type but also on when it is sent. One million uncached input tokens cost 1 yuan off-peak and output costs 4 yuan, compared with 2 yuan and 8 yuan during peak hours. Batch evaluation, indexing or agent work that is not latency-sensitive could therefore benefit if it can be scheduled outside peak periods. The announcement does not say how many tokens a model will consume for a task or how many attempts it will need, so a lower unit price does not mean the same proportional reduction in total cost for every job.[1]
