Hangzhou-based AI startup DeepSeek has released a preview of its latest large language model, V4, intensifying the global competition in artificial intelligence. The company, which shook the industry in early 2025 with models that delivered strong performance at a fraction of the cost of US rivals, now offers two versions: V4-Pro for demanding tasks and V4-Flash for faster, cheaper responses.
DeepSeek claims that V4-Pro “significantly leads other open source models” in world knowledge benchmarks and is “only slightly outperformed by the top-tier closed-source model Gemini-3.1-Pro” from Google. The new models support a “one-million token context length,” allowing them to process large documents or code before generating answers. In a post on Hugging Face, the company declared: “Welcome to the era of cost-effective 1M context length.”
The release comes just a day after OpenAI launched GPT-5.5, which it called its “smartest and most intuitive” model yet. It also follows accusations from the White House that Chinese firms, including DeepSeek, are stealing American AI labs’ intellectual property on an industrial scale using thousands of proxy accounts. OpenAI, Anthropic, and Google have warned that Chinese AI companies are conducting “model extraction attacks,” or distillation, where a larger model is fed thousands of questions and its answers are used to train a smaller, cheaper model.
European Concerns Over Data Security
DeepSeek’s rise has not gone unnoticed by European regulators. In 2025, Italy, Germany, and South Korea banned government agencies from using DeepSeek over national security concerns. Germany also removed the app from Apple and Google stores, citing illegal transfer of user data to China. These actions reflect broader European anxieties about data sovereignty and the geopolitical implications of AI development.
Despite these tensions, DeepSeek has positioned itself as a champion of open-source AI. The company has made V4 available for download on Hugging Face, allowing developers to test and adapt the model freely. It can also be used with popular AI agents like Claude Code, OpenClaw, and OpenCode, enabling software tasks beyond DeepSeek’s own chatbot. This openness contrasts sharply with the closed models of US rivals like OpenAI and Anthropic.
The startup first gained attention in late 2024 with its V3 model, which surprised the industry by delivering strong performance using less powerful chips and at lower cost than many US competitors. In January 2025, its R1 reasoning model claimed to match OpenAI’s ChatGPT at a lower price, further cementing its reputation as a cost-effective alternative.
As the AI race accelerates, European policymakers face a delicate balancing act: fostering innovation while protecting citizens’ data and maintaining strategic autonomy. The release of DeepSeek V4, alongside GPT-5.5, underscores the urgency of developing clear, enforceable rules for AI governance across the continent.
