The Chinese startup DeepSeek is preparing to unveil its next-generation model, V4, focused on programming, which could surpass Western AI systems.

The launch window is estimated in the coming weeks, possibly around the Lunar New Year in February. Internal tests suggest that V4 will emphasize advanced coding capabilities and could surpass OpenAI and Anthropic systems in programming-related tasks. This release is strategic, considering the previous success of the R1 model, launched a few days before the Lunar New Year, which generated intense discussions in technology and financial circles. DeepSeek, supported by the High-Flyer Quant hedge fund, has gained popularity not only due to the quality of the model but also its efficiency, with R1 achieving remarkable reasoning performance with reduced computational resources. The V3.2 model has already demonstrated superior performance compared to other top models, but V4 is expected to be a significant architectural leap. The focus on coding is crucial, as software development represents a valuable use case for AI in business. If V4 lives up to its promises, DeepSeek could position itself as a high-performance alternative with lower costs compared to American models. The launch adds uncertainty to the market, considering that R1 has influenced the actions of American chip companies and AI leaders, demonstrating that models outside Silicon Valley can change narratives about dominance and capital advantage.