DeepSeek launches V4 Pro model, increases API price 11 times

On August 13, DeepSeek officially announced version V4-Pro-0813 with a major upgrade in AI Agent processing capabilities. This model is currently open for simultaneous access on application platforms, web interfaces and API connection portals.

According to Reuters, the launch of V4 Pro attracted great attention from the technology world. Previously, the low-cost version V4 Flash unexpectedly surpassed the performance of the V4 Pro test version in April in many independent tests. The new move shows that DeepSeek has made technological advances in a short time to complete the official version of V4 Pro.

The company also announced to adjust API service prices for both V4 Pro and V4 Flash lines, and for the first time applied a price division mechanism according to peak and off-peak hours. The new price is expected to increase from 50% to 1,100% (11 times) depending on the type of model, token, time frame of use and will take effect from August 17.

To maintain competitive advantage, Chinese startups are also transforming from a pure research unit to a large-scale business model. According to Reuters, DeepSeek is preparing for a new round of capital raising with a valuation of $74 billion, a few weeks after successfully raising $7.4 billion in its first external capital round in June.

Due to the huge cost of maintaining AI infrastructure, including data centers, chip procurement and talent recruitment, DeepSeek plans to double its staff size, especially in the areas of data center operations and AI Agent development. The company is also recruiting chip design engineers to develop its own line of AI processors, aiming to reduce dependence on hardware suppliers such as Nvidia and Huawei.

After causing a global sensation in early 2025 with the R1 model, DeepSeek is facing fierce competition right in the domestic market from competitors such as Moonshot AI, Zhipu AI, MiniMax, Alibaba or ByteDance.

Huy Duc