DeepSeek has released a preview of its V4 AI model, featuring a 1 million token context window and support for Huawei's domestic AI chips. This open-source initiative includes two versions: V4 Pro, with 1.6 trillion total parameters, designed for complex tasks; and V4 Flash, with 284 billion total parameters, offering a faster, more economical option.
This launch marks DeepSeek's first major model adapted for Huawei hardware, utilizing Ascend 950 AI chips. This collaboration aligns with China's drive to reduce dependency on foreign AI infrastructure, with Huawei's hardware also supporting the training of the V4 Flash version.
Previous DeepSeek models relied on Nvidia chips, but U.S. export restrictions have prompted Chinese AI developers to adopt domestic alternatives. DeepSeek's V4 aims to challenge U.S. AI companies by offering competitive performance while remaining open source, echoing its past success with the R1 model.