DeepSeek reportedly did not share its upcoming AI model with American developers and instead gave early access to Chinese companies, further intensifying the technological war between the US and China, since Feb. 26, 2026.
Nurphoto Nurphoto Getty Images
Chinese Artificial Intelligence DeepSeek on Friday released a preview version of its long-awaited V4 major language model, allowing users to test its new capabilities and features.
The release comes more than a year after DeepSeek introduced its R1 imaging model, which rocked global technology markets with its impressive performance and cost-effectiveness.
Similar to DeepSeek’s previous V3 model, the latest development is open source, allowing developers to download the code, run it locally, and modify it.
The Hangzhou-based company said the V4 achieves strong performance compared to domestic rivals, especially in agent-based tasks, information processing and prediction.
The company added that DeepSeek-V4 is optimized for use with popular agent tools such as Anthropic’s Claude Code and OpenClaw.
The model is available in both “pro” and “flash” versions, depending on the size.
Founded in 2023, DeepSeek gained attention in late 2024 for its free, open-source V3 model, which it claims was trained with less powerful chips and at a fraction of the cost of models built by the likes of OpenAI and Google.
Weeks later, in January 2025, it released a conceptual model, R1, which achieved the same ratings or surpassed many of the world’s leading LLMs.
The R1 model shocked investors when DeepSeek revealed that it took only two months, not even $6 million, to build this model using low-end Nvidia chips. That called into question US leadership in AI and Big Tech’s heavy spending on AI infrastructure.
Since then, DeepSeek has released a series of model upgrades, but nothing quite as impactful as the R1.
V4 is also unlikely to have the same impact as R1, because markets have already priced in the fact that Chinese AI is competitive and cheap to use, Ivan Su, senior financial analyst at MorningStar, told CNBC.
However, DeepSeek’s recent stance puts other Chinese open source models as direct competitors, Su said.
“This is a structure that was not there in R1, and that alone tells you how intense the domestic competition is,” he added.
Since the release of R1, DeepSeek has faced increased competition in China’s booming AI sector, with players like Alibaba and ByteDance is also releasing new models this year.
Shares of several other Chinese AI players were lower in Hong Kong trading on Friday. MiniMax and Knowledge Atlas Technology, also known as Zipu, each fell about 8%, while Hangzhou-based developer Manycore Tech fell 9%.
Which chips train the V4?
A big question following the release of DeepSeek’s V4 model is which chips were used to train the model.
Chinese tech giant Huawei on Friday confirmed that its latest batch of AI computers, powered by Ascend AI processors, can support DeepSeek’s V4 model.
However, it is unclear how much Huawei’s Ascend chips were used in model training compared to Nvidia’s.
Chinese AI developers are barred from directly buying Nvidia’s most advanced chips for AI training because of US export controls.
Meanwhile, Beijing has stepped up efforts to develop its domestic chip industry and reportedly pressured Chinese tech companies to find domestic alternatives from chipmakers such as Huawei, rather than foreign alternatives.
After DeepSeek announced its V4 release, shares of the Chinese contract chipmaker rose in Hong Kong, with SMIC again Hua Hong Semiconductor increased by 8.9% and 15.2%, respectively.
