Introduction#
On Friday, DeepSeek announced the release of its new flagship open source artificial intelligence model, V4. This model is designed to enhance reasoning and performance capabilities, marking a significant advancement in the field of AI.
New Model Versions#
DeepSeek has introduced two versions of the V4 model: DeepSeek-V4-Pro and DeepSeek-V4-Flash. The Pro version is equipped with 1.6 trillion parameters, which are the individual elements that allow the model to learn and process information. In contrast, the Flash version is a more compact model with 284 billion parameters, making it a leaner option for users.
Performance Highlights#
According to DeepSeek, the V4-Pro-Max, the most advanced model in this release, has achieved impressive results in coding benchmarks. It is reported to perform comparably to leading closed-source models, particularly in reasoning tasks. On the MMLU-Pro benchmark, which is widely used to assess AI coding capabilities, DeepSeek V4-Pro matched the performance of OpenAI’s GPT-5.4, while slightly lagging behind Google’s Gemini-3.1-Pro and Anthropic’s Claude Opus 4.6.
Industry Impact#
This announcement marks DeepSeek's first major model release since its R1 model made waves in early 2025. The R1 model was seen as a significant breakthrough for open source AI, offering performance levels that rivaled those of closed-source competitors. The initial release of R1 had a notable impact on global technology stocks, as investors began to question the necessity for additional AI infrastructure when a more efficient model could deliver similar results.
Additionally, this release comes shortly after reports indicated that Chinese tech giants Tencent and Alibaba are in discussions to invest in DeepSeek, valuing the company at over $20 billion. DeepSeek is recognized as one of China’s leading AI companies, part of a group known as the "AI Tigers," which are pivotal to the country's ambitions in the rapidly evolving AI industry.