DeepSeek Slashes Prices for Top Model V4-Pro by 75 Percent, Renews Attack on Anthropic and Co.

Chinese AI startup DeepSeek has made permanent the price cut for its flagship model V4-Pro that it introduced last month. API access costs now remain at just a quarter of their original level -- an aggressive move amid an escalating global price war for enterprise customers in the large language model segment.

Specifically, V4-Pro API prices drop to a range between 0.025 and 6 yuan per million tokens, equivalent to roughly 0.0035 to 0.83 US dollars per million units. Before the adjustment, rates ranged from 0.1 to 24 yuan per million tokens. The exact pricing depends on the type of usage -- cheaper for pure text input, more expensive for the compute-intensive text generation. A token refers to the smallest semantic unit an algorithm can process, typically a word or part of a word.

With this pricing structure, DeepSeek positions itself internationally as one of the cheapest providers in the premium segment for language models. The move ramps up pressure on Western market leaders such as OpenAI and Anthropic, whose top-tier models typically cost multiples of that per million tokens. Chinese providers are increasingly trying to use aggressive pricing to cut directly into the margin structure of their American competitors.

The pricing offensive is tightly linked to DeepSeek's hardware strategy: for V4, the company relies on Huawei Ascend 950 semiconductors rather than Nvidia GPUs, which remain difficult to access for Chinese customers due to US export controls. When V4 launched last month, DeepSeek had priced the Pro version up to twelve times higher than the leaner Flash variant, citing bottlenecks in high-end compute capacity.

Whether the permanent price reduction is directly attributable to an easing Huawei supply chain, the company left open. However, DeepSeek had previously predicted that infrastructure costs would fall significantly once the so-called supernodes of the Ascend series ship in larger volumes in the second half of 2026. The combination of DeepSeek and Huawei is strategically seen as the core of an independent Chinese AI stack designed to reduce dependence on US semiconductors.

Alongside the pricing offensive, DeepSeek is preparing its first-ever external funding round. The lab founded by hedge fund billionaire Liang Wenfeng -- who controls roughly 90 percent of the equity -- has so far consistently avoided outside capital, financing itself through Liang's hedge fund High-Flyer. Now the company is reportedly targeting a round of three to four billion US dollars, according to Reuters and other outlets.

The valuation has climbed rapidly within just a few weeks: DeepSeek initially sought a valuation of at least 10 billion dollars when talks began, with the figure rising to around 20 billion by late April. Recent reports from the Financial Times and Bloomberg now cite 45 billion dollars, while the South China Morning Post even mentions up to 50 billion.

According to the reports, the round is being led by China's state-backed semiconductor investment vehicle, the China Integrated Circuit Industry Investment Fund ("Big Fund III"). Tencent, Alibaba, and investor Hillhouse are also said to be participating. If the deal closes, it would mark a milestone: it would be the Big Fund's first known investment in a Chinese large language model company -- a clear political signal that Beijing is positioning DeepSeek as a national champion in the AI race with the United States.

According to the Financial Times, Liang decided to open the cap table primarily to be able to offer employees equity -- the background being increasing poaching by competitors such as Alibaba, ByteDance, and Moonshot AI. The fresh capital is intended to flow primarily into expanding the compute infrastructure.