DeepSeek releases V4 preview as Chinese AI competition intensifies
Chinese AI startup DeepSeek released a preview of its V4 large language model on Friday, expanding its open-source offerings and signaling continued momentum in a crowded domestic market.
The Hangzhou-based company introduced V4 more than a year after its R1 reasoning model disrupted global markets by matching or exceeding competitors' performance at a fraction of the cost. The new model comes in "pro" and "flash" versions, with DeepSeek claiming stronger performance in agent-based tasks, knowledge processing, and inference.
DeepSeek has optimized V4 to work with popular agent tools, including Anthropic's Claude Code and OpenClaw. The model is open-source, allowing developers to download, run locally, and modify the code in most cases.
Lower costs, but diminished shock value
Analysts expect V4 to offer lower inference costs than previous models. Wei Sun, principal AI analyst at Counterpoint Research, said the benchmarks suggest "excellent agent capability at significantly lower cost."
However, V4 is unlikely to produce the market impact of R1. When DeepSeek revealed in January 2025 that R1 took only two months and under $6 million to build using lower-capacity Nvidia chips, it rattled investors and raised questions about U.S. AI dominance. Markets have since absorbed the reality that Chinese AI is competitive and cheaper.
Ivan Su, senior equity analyst at Morningstar, noted that V4's real significance lies elsewhere: it positions other Chinese models as direct competitors to DeepSeek. "This is a framing that didn't exist with R1, and that alone tells you how much domestic competition has intensified," he said.
Since R1's release, Alibaba and ByteDance have launched competing models. On Friday, shares of Chinese AI companies MiniMax and Zhipu each fell around 8%, while Manycore Tech dropped 9%.
The chip question
A central question remains: which chips trained V4? Huawei confirmed Friday that its Ascend AI processors can support the model, but the extent of their use in training versus Nvidia chips is unclear.
U.S. export controls restrict Chinese developers from purchasing Nvidia's most advanced AI chips. Beijing has pushed domestic tech companies toward alternatives from chipmakers like Huawei.
If V4 runs natively on local chips, the implications are significant for China's AI sovereignty and reduced reliance on Nvidia. Wei Sun said this could "speed up global AI developments as well."
Markets responded accordingly. Chinese contract chip manufacturers surged in Hong Kong trading: SMIC rose 9% and Hua Hong Semiconductor jumped 15%.
For product and development teams, V4's release reinforces a shift in the competitive landscape. Generative AI and LLM capabilities are now available across multiple vendors at lower costs, changing the calculus for build-versus-buy decisions and infrastructure spending.
Your membership also unlocks:
