On the evening of June 29, many developers reported receiving emails stating that DeepSeek would launch the official version of V4 and adjust billing. The email mentioned that the official version of DeepSeek V4 is scheduled to be officially launched in mid-July, which will bring more feature optimizations and performance improvements. At the same time, DeepSeek will simultaneously adjust its API pricing strategy and introduce a peak and valley pricing mechanism "to allocate resources more reasonably and improve service stability."
China Business News reporter saw on the DeepSeek open platform that the official also sent a notice to users on the homepage, saying that the API service is expected to adopt a peak and valley pricing strategy from mid-July, and "the price during peak hours will be twice the usual price."
Specifically, peak hours are defined as 9:00-12:00 and 14:00-18:00 Beijing time every day, which almost covers the most important working hours of domestic developers and enterprise users.
In terms of price, the V4 Pro cache hit input price during peak hours is 0.05 yuan/million Tokens, the cache miss input price is 6 yuan/million Tokens, and the output price is 12 yuan/million Tokens. During peak hours of V4 Flash, the cache hit input price is 0.04 yuan/million Tokens, the cache miss input price is 2 yuan/million Tokens, and the output price is 4 yuan/million Tokens.
Previously, on April 24, DeepSeek announced that the V4 preview version was officially launched and open sourced simultaneously. Since then, there have been several price adjustments. On May 22, DeepSeek announced that it would adjust the original limited-time 25% discount of the DeepSeek-V4 model to a permanent price reduction. The discounted preview version price is consistent with the off-peak price announced this time.
This means that the official version of DeepSeek V4 has not increased in overall price, but if users choose to call the model API during peak hours, the cost will directly double.
After the news was released, it attracted heated discussions among many developers. Some developers believe that this time division is not friendly to domestic users, but for users in time zones such as Europe and the United States, most of the call times fall in valley hours, which to a certain extent is more like "benefiting overseas users."
Many developers also expressed understanding. A developer who has used the DeepSeek API for a long time told reporters that DeepSeek's current pricing itself is already at a low level in the industry. Even if the price increases to 2 times during peak hours, the overall cost is still competitive. "The real key is not the price, but the capability upgrade of the official version of V4."
What more users are looking forward to is whether the official version of DeepSeek V4 can enhance its performance in post-training, multi-modality, programming, Agent, etc. while the peak and valley pricing is in place.
From an industry perspective, peak and valley pricing may mean that AI cloud services are maturing. Some developers mentioned that computing power is no longer unified at a fixed price, and the cost of tokens will be closer to the electricity price model in the future. In the future, other large domestic models will most likely follow the flexible pricing, and tiered billing of computing power by time will become the norm in the industry, but this does increase scheduling costs for small and medium-sized developers.
This also means that the usage habits of developers will be reshaped in the future. Peak and valley pricing will force developers to do caching and task scheduling. More batch tasks may be scheduled to run during off-peak periods. Task orchestration capabilities will become a new competitiveness for developers to control costs.