On August 13, DeepSeek unveiled the official release of its V4 Pro model, deploying updates simultaneously across its mobile application, web interface, and API platforms. Users can now tap into the advanced capabilities of this new iteration through the "Expert Mode" feature on both the app and web versions, while the API model retains its original naming convention. One of the standout features of V4 Pro is its native support for ultra-long contexts, accommodating up to a staggering 1 million tokens, along with a maximum single output capacity of 384,000 tokens. The model's Agent functionalities have undergone substantial enhancements, showcasing particularly impressive performance in real-world production settings. Moreover, the thinking modes of both V4-Pro and V4-Flash now offer three distinct levels of thinking intensity—low, high, and max—empowering users to tailor their experience according to the complexity of their tasks. In addition to these technical upgrades, DeepSeek API is set to implement a peak-valley pricing strategy, with the new pricing structure coming into effect at midnight on August 17.