Key Highlights
- DeepSeek unveiled two open source AI models on Friday: V4-Pro (featuring 1.6T parameters) and V4-Flash (with 284B parameters)
- Each model offers a one million token context window, comparable to capabilities found in Google’s Gemini
- V4-Pro achieves performance parity with OpenAI’s GPT-5.4 in coding tests and ranks second only to Gemini in reasoning tasks
- The company emphasizes significantly lower computational and memory requirements versus competitors
- This launch follows news that Tencent and Alibaba are considering investments in DeepSeek at valuations exceeding $20 billion
On Friday, Chinese artificial intelligence company DeepSeek introduced preview editions of V4, its latest flagship open source model. The firm claims the new release delivers enhanced reasoning capabilities, reduced operational expenses, and an expansive context processing capacity.
The launch includes two distinct variants: V4-Pro and V4-Flash. The Pro edition features 1.6 trillion parameters, while Flash represents a streamlined alternative with 284 billion parameters, engineered for superior efficiency and cost-effectiveness.
Both variants accommodate a context window spanning one million tokens. This capability enables them to analyze substantial volumes of text simultaneously, positioning them alongside Google’s Gemini in this dimension.
DeepSeek noted that current models process text exclusively. The organization is developing multimodal functionality that will enable future versions to handle images and video content.
Performance Against Leading Competitors
On the MMLU-Pro benchmark, a standard AI evaluation metric, V4-Pro delivered results equivalent to OpenAI’s GPT-5.4. It scored marginally below Google’s Gemini and Anthropic’s Claude Opus 4.6. For reasoning assessments, V4-Pro secured second place, trailing only Google’s most recent Gemini release.
DeepSeek further indicated that V4 has been fine-tuned for AI agent platforms such as Claude Code, OpenCode, and CodeBuddy.
The organization characterized V4’s context capacity as “world leading with drastically reduced compute and memory costs.” Industry analyst Zhang Yi labeled it a transformative moment, suggesting that ultra-extensive context capabilities could transition from experimental settings into mainstream commercial applications.
AI researcher Max Liu described the announcement as a “milestone” for China’s artificial intelligence sector, drawing parallels to the market impact of DeepSeek’s original R1 model launch.
Investment Landscape and Market Implications
This represents DeepSeek’s first comprehensive model architecture since R1 debuted in early 2025. That earlier release disrupted global technology markets, affecting companies like Nvidia and Meta by demonstrating that cost-efficient, streamlined models could rival expensive proprietary alternatives.
DeepSeek has not disclosed which processors were utilized for V4 training. Earlier in 2025, U.S. authorities alleged the company employed restricted Nvidia Blackwell chips. According to The Information, the models were actually trained using Huawei processors.
Huawei verified that its Ascend supernode infrastructure, powered by Ascend 950 AI chips, offers complete compatibility with DeepSeek’s V4 models.
The announcement arrives shortly after media reports indicated that Tencent and Alibaba are pursuing investment discussions with DeepSeek at valuations surpassing $20 billion. DeepSeek ranks among China’s top six AI unicorn companies.
A preview build of V4 is currently accessible through Hugging Face. DeepSeek has not provided a timeline for the complete public release.
