Chinese artificial intelligence startup DeepSeek unveiled a preview of its fourth-generation model, DeepSeek V4, on Friday, marking its most significant product update since the V3 release in December 2024. The new model, available in pro and flash versions, is positioned as a major upgrade and successor to V3, which propelled the company into the global spotlight for its strong performance at low cost.
The launch comes at a critical time in the AI race, with investors closely monitoring Chinese model makers. DeepSeek described the preview as an early-stage rollout aimed at gathering real-world feedback before wider deployment, signaling a measured approach to product testing and refinement. The company did not disclose pricing, performance benchmarks, or a final release timetable.
DeepSeek V4 is open-source and designed to work with popular agent tools, including Anthropic's Claude Code. The model performs well against domestic rivals in AI agents, knowledge handling, and inference tasks, according to the company. This release follows DeepSeek's R1 reasoning model from January 2025, which shocked global markets by matching leading AI models while costing less than $6 million to build using lower-grade Nvidia chips.
Market Reaction: AI Stocks Fall, Chipmakers Surge
The V4 preview sent shares of several Chinese AI companies lower in Hong Kong. Zhipu AI fell approximately 8-9%, MiniMax dropped 7-8%, and Manycore Tech slid 9%. Conversely, chipmaking stocks rallied on optimism over AI-driven demand. Semiconductor Manufacturing International Corp (SMIC) jumped 11%, Hua Hong Semiconductor rallied over 18%, and on the mainland, Cambricon Technologies and Moore Threads Technology gained 4-6%, while Hygon Information Technology climbed over 10%.
Hardware and Export Control Questions
One of the biggest questions surrounding V4 is which chips were used for training. Reuters reported that Huawei confirmed its Ascend 950-based supernode supports the V4 model, but DeepSeek did not disclose its training hardware. Chinese AI developers have been blocked from buying Nvidia's most advanced chips due to U.S. export controls since 2022, prompting a push towards domestic alternatives like Huawei's chips. The launch also comes a day after the White House accused China of stealing U.S. AI intellectual property, a charge the Chinese Embassy in Washington dismissed as baseless.
First Outside Fundraising for DeepSeek
DeepSeek is also in talks to raise outside funding for the first time, targeting a valuation above $20 billion. The fundraising, expected to be in the low hundreds of millions of dollars, is primarily aimed at retaining researchers who have left for rivals like ByteDance and Tencent, as stock options form a large part of AI researcher compensation. Founder Liang Wenfeng, who has funded the company through his quantitative trading firm, is also considering alternative valuation methods such as a share buyback.
