DeepSeek Unveils New V4-Pro and V4-Flash Open-Source AI Models with 1 Million Token Context

Chinese AI startup DeepSeek has unveiled preview versions of its new flagship open-source models, V4-Pro and V4-Flash, marking a significant step in democratizing high-performance AI. The V4-Pro model boasts a massive 1.6 trillion total parameters, though only 49 billion are activated per inference, which helps balance performance with computational efficiency. Both models natively support a 1 million-token context window, enabling them to process entire books or extensive codebases in a single pass. DeepSeek claims the V4-Pro matches or exceeds the performance of leading closed-source models like GPT-4 and Claude 3.5 on coding and agentic tasks, such as automated code generation and multi-step reasoning.

Both models are released under the permissive MIT license, with weights available on Hugging Face, allowing unrestricted use, modification, and commercial deployment. DeepSeek is also offering aggressive API pricing, with rates significantly lower than comparable closed-source alternatives, to encourage widespread adoption. The V4-Flash variant, while slightly less capable, is optimized for speed and cost efficiency, making it suitable for real-time applications. This launch positions DeepSeek as a serious competitor in the open-source AI space, challenging the dominance of US-based models and providing developers with powerful, affordable tools for building next-generation applications.