DeepSeek Unveils V4: The Latest Open-Source AI Model Challenging Big Tech Giants

Key Highlights

• DeepSeek unveiled two open-source AI models: V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion parameters)
• Each model features a 1-million-token context window, rivaling Google’s Gemini capabilities
• V4-Pro achieves performance comparable to 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
• News arrives amid reports of Tencent and Alibaba negotiating investment deals valuing DeepSeek above $20 billion

Chinese artificial intelligence firm DeepSeek unveiled preview editions of its newest flagship open-source AI system, V4, this past Friday. According to the company, this latest iteration delivers enhanced reasoning capabilities, cost efficiency, and an exceptionally large context processing capacity.

The firm introduced two distinct variants: V4-Pro and V4-Flash. The Pro edition features 1.6 trillion parameters, while the Flash variant represents a streamlined alternative containing 284 billion parameters, engineered for superior efficiency and cost-effectiveness.

Each variant supports processing up to one million tokens simultaneously. This capability enables them to analyze substantial volumes of text in a single operation, positioning them competitively alongside Google’s Gemini in this dimension.

The company noted that current models handle text exclusively. DeepSeek confirmed development is underway to incorporate multimodal functionality, which will enable future versions to analyze images and video content.

Performance Against Competing Systems

In MMLU-Pro testing, a standard industry benchmark, V4-Pro delivered results equivalent to OpenAI’s GPT-5.4. Performance placed it marginally below Google’s Gemini and Anthropic’s Claude Opus 4.6. For reasoning benchmarks specifically, V4-Pro secured second place behind only the most recent Gemini release.

DeepSeek highlighted that V4 has been fine-tuned for integration with AI agent frameworks including 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 identified it as an “inflection point,” suggesting ultra-long context capabilities could transition from experimental research environments into mainstream commercial applications.

AI industry expert Max Liu characterized the launch as a “milestone” for China’s artificial intelligence sector, drawing parallels to the market impact when DeepSeek’s R1 initially debuted.

Financial and Strategic Landscape

This marks DeepSeek’s first significant new-generation model launch since R1 emerged in early 2025. That previous release sent ripples through global technology markets, affecting companies like Nvidia and Meta, by demonstrating that an economical, efficient model could rival expensive proprietary alternatives.

DeepSeek has not disclosed which semiconductor chips powered V4’s training process. Earlier in the year, U.S. authorities alleged the company utilized restricted Nvidia Blackwell chips. Subsequently, a report from The Information indicated training occurred on Huawei chips instead.

Huawei verified that its Ascend supernode infrastructure, utilizing Ascend 950 AI processors, would provide complete support for DeepSeek’s V4 systems.

The model debut follows closely after reports emerged that Tencent and Alibaba are pursuing investment discussions with DeepSeek at a valuation exceeding $20 billion. DeepSeek ranks among China’s six premier AI unicorn companies.

A preview build of V4 is currently accessible through Hugging Face. DeepSeek has not yet specified a timeline for the complete public release.