DeepSeek V4: A New Contender in the AI Model Race
Chinese startup DeepSeek has rolled out preview versions of its much-anticipated V4 model, immediately drawing comparisons with leading US systems such as OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini. The new DeepSeek AI model comes in “pro” and “flash” variants and is positioned as open-source or open-style, with major gains in knowledge, reasoning and so‑called “agentic” capabilities – the ability to autonomously execute complex workflows. DeepSeek claims its V4 Pro Max surpasses OpenAI’s GPT‑5.2 and Google’s Gemini 3.0‑Pro on standard reasoning benchmarks, while falling only marginally behind GPT‑5.4 and Gemini 3.1‑Pro. The company also highlights a one‑million‑token context window for both “pro” and “flash”, enabling long documents and multi‑step tasks. Notably, V4 is partly powered by Huawei chips, reducing reliance on US suppliers such as Nvidia and underlining China’s push for self‑reliance in critical AI compute.
Open-Style AI: Why DeepSeek Appeals to Global Developers
DeepSeek promotes V4 as an open-source AI family that lets developers modify and build on its core models, in contrast to the tightly controlled, closed platforms of most US rivals. Beyond a free‑to‑use web and mobile chatbot, the company’s open-style approach promises access to weights and tools that make fine‑tuning, customization and on-premises deployment easier for startups and enterprises. For global AI builders, especially those outside the US tech ecosystem, this combination of high benchmark performance, agentic capabilities and flexible licensing is attractive. It offers a way to experiment with cutting‑edge AI without being locked into a single US platform’s pricing, policy changes or data‑use restrictions. That appeal is amplified in emerging markets, where access to powerful yet affordable models can determine whether local companies can compete in areas like automation, customer service, and AI‑augmented software products.
US Crackdown on AI Model Distillation and IP Theft Concerns
Washington is responding to this wave of open‑ish Chinese models with growing alarm. A recent US State Department directive instructs diplomats worldwide to warn governments about what it calls “industrial‑scale” AI model distillation by Chinese firms, explicitly naming high‑profile players like DeepSeek. Distillation itself – training smaller models using outputs from more powerful systems – is a common technique. But US officials allege that Chinese actors are stripping safety layers and effectively cloning proprietary US models at a fraction of the R&D cost, turning AI geopolitics into a battle over intellectual property and national security. The diplomatic cable lays the groundwork for possible follow‑up measures, from tighter export controls on advanced AI systems and chips to new restrictions on cross‑border access to US models. Beijing has dismissed the allegations as slander, insisting it respects IP rights, while DeepSeek continues to deny any wrongdoing.
DeepSeek as China’s AI ‘Darling’ Amid Rising US–China AI Tensions
DeepSeek has quickly become a symbol of China’s AI ambitions and the broader US China AI tensions. Its earlier low‑cost reasoning model R1 shocked markets and showcased how quickly Chinese labs were closing the gap with US leaders. Reports suggest DeepSeek is now seeking funds at a valuation exceeding tens of billions and attracting interest from Chinese tech giants such as Alibaba and Tencent. At the same time, US rivals and officials accuse the company of improper behaviour, including building on US models without authorization and benefitting from the alleged AI model distillation of American systems. The White House recently charged that China was stealing AI lab IP on an industrial scale, a claim Beijing rejected as baseless. DeepSeek’s V4 launch, powered partly by Huawei chips and framed as a milestone of technological self‑reliance, reinforces its status as China’s AI “darling” – and a lightning rod in AI geopolitics.
Implications for Developers and Malaysian Companies in a Fracturing AI Market
For international developers, especially in Southeast Asia, DeepSeek’s open-style models offer both opportunity and risk. On one hand, competitive performance, open access and lower‑cost variants like V4 “flash” can accelerate product development across sectors from fintech to logistics. On the other, intensifying US China AI tensions raise questions about future sanctions, data‑transfer limits and access to updates or APIs. Malaysian companies need to treat AI procurement like any other strategic technology in a geopolitically sensitive supply chain. Practical steps include: assessing vendor risk where models or infrastructure originate in jurisdictions exposed to sanctions; ensuring data residency and compliance with local regulations when using foreign AI; and diversifying AI model suppliers, mixing US, Chinese and neutral providers or self‑hosted open source AI where possible. Building internal capabilities to swap or layer models will be critical as global AI ecosystems fragment along geopolitical lines.
