DeepSeek, the Chinese artificial intelligence startup that disrupted global markets last year, has released preview versions of its latest major update, intensifying the AI competition between China and the United States.
DeepSeek V4 Launches with Enhanced Capabilities
The new V4 open-source models, including "pro" and "flash" versions, boast significant improvements in knowledge, reasoning, and "agentic" capabilities—the ability to perform complex tasks autonomously. DeepSeek claims the V4 Pro Max version offers "superior performance" on standard reasoning benchmarks compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, though it falls "marginally" short of GPT-5.4 and Gemini 3.1-Pro.
Agentic Performance and Comparisons
In agentic capabilities, DeepSeek says the V4 "pro" version can outperform Claude's Sonnet 4.5 and approaches the level of Claude's Opus 4.5 based on its own evaluation. The "flash" version performs on par with the "pro" on simple agent tasks and has closely approaching reasoning capabilities.
Market Expectations and Context
Industry analysts had anticipated the model's release more than a month earlier, around the Lunar New Year. DeepSeek's specialized reasoning model, R1, released in January 2025, had previously surprised markets by being more cost-effective than OpenAI's similar model, symbolizing China's technological catch-up.
Expert Opinions
Lian Jye Su, chief analyst at Omdia, noted, "Based on the benchmark results, it does appear DeepSeek V4 is going to be very competitive against its U.S. rivals." Marina Zhang, associate professor at the University of Technology Sydney, called the rollout a "pivotal milestone for China's AI industry" amid global competition for self-reliance in critical technologies.
Open Source and Technical Details
DeepSeek offers a free-to-use web and mobile chatbot and describes its technology as "open source," allowing developers to modify and build on its core technology. Both V4 versions have a 1 million token context window—a significant increase from V3's 128,000 tokens—and run more efficiently.
Allegations of Distillation
Anthropic and OpenAI have accused DeepSeek of using distillation—training a weaker model on a stronger one's outputs—to illicitly extract capabilities. In February, Anthropic accused DeepSeek and two other Chinese labs of "industrial-scale campaigns" to do so. OpenAI made similar allegations to U.S. lawmakers. This week, U.S. Chief Science and Technology Adviser Michael Kratsios accused foreign tech companies "principally based in China" of distilling leading U.S. AI systems. China's embassy in Washington dismissed the allegations as "unjustified suppression."
Market Adoption and Skepticism
A Microsoft report from January showed DeepSeek gaining ground in developing nations. However, Morningstar senior equity analyst Ivan Su cautioned that while V4 is a "competent" follow-up, it is not as groundbreaking as R1. He noted that independent evaluations are needed before final conclusions can be drawn.
