DeepSeek has unveiled its new AI model, V4, as one of its chief initiatives to try to establish itself as the leading global player in artificial intelligence.
V4 is an open-source model that is designed to compete with the best closed-source systems developed by companies such as OpenAI and Google DeepMind.
The company released two different versions of V4. The V4-pro version has 1.6 trillion parameters, the most of any model produced by DeepSeek to date. The other version, V4-flash, has 284 billion parameters.
Both V4 models have implemented a new maximum context window size of 1 million tokens (far larger than the previous 128,000), which gives them a better ability to process and retain information during a conversation.
DeepSeek claims it achieved this with strong cost efficiency, which could matter just as much as raw performance in real-world use.
Huawei steps in for DeepSeek Support
Shortly after the launch, Huawei announced full support for the new models through its Ascend chip lineup and supernode systems.
This backing is significant. It signals a deeper push toward building a domestic AI ecosystem that relies less on foreign hardware. Chipmaker Cambricon Technologies also confirmed compatibility, adding further momentum.
Analysts say this could accelerate the adoption of Chinese-made AI chips and improve their capabilities faster than expected.
Performance comes with trade-offs
Although V4 pro is pushing the envelope in regard to the features and the size of the hardware it can run on, you will need to invest in more than just typical hardware to run V4 pro.
The demands of the processor that run V4 pro are higher than anything we’ve seen before, and, according to DeepSeek, current performance is limited by the amount of available hardware.
According to DeepSeek, they expect the price of V4 pro to drop significantly in late 2022 when Huawei’s new Ascend 950 PR systems are introduced to the market.
On the flip side of that is V4 flash, whose goal is to be a commercially available model and is expected to be one of the least expensive state-of-the-art AI models on the market.
Despite all of the talk about V4, it continues to be a text model, and according to DeepSeek, they are working on building additional features using images and video into the V4 model, but those features will not be released until a later version of V4.
The ripple effects of the announcement regarding V4 are already being felt well beyond the software sphere.
Major Chinese semiconductor manufacturers have seen significant increases in share price in response to the launch of V4, as a direct result of the connection between advancements in the field of AI and the semiconductor industry, especially within China, where export restrictions by the US are adversely affecting the growth of AI and semiconductor technologies in China.
DeepSeek’s V4 isn’t just another model drop. It reflects a broader shift. China’s AI companies are not only trying to catch up with global leaders but also building their own ecosystem across software and hardware.
With Huawei stepping in and domestic chip support growing, the gap may not stay as wide for long.
Also Read: DeepSeek V3.2-Speciale Challenges Google Gemini 3 Pro In Global AI Race
