DeepSeek is poised to unveil its latest multimodal large language model, V4, next week. This new model will inherently support the generation of images, videos, and text. This release signifies DeepSeek's first significant version upgrade since the introduction of the R1 reasoning model in January 2025. It is anticipated to fulfill the domestic market's need for cost-effective, open-source models. In conjunction with the launch of V4, DeepSeek will issue a concise technical overview and subsequently release a comprehensive engineering report, roughly one month thereafter. Additionally, DeepSeek is offering extensive support for domestic computing capabilities, having worked in tandem with Huawei and Cambrian to enhance hardware compatibility. This move is expected to spur growth in the local semiconductor sector and promote the utilization of domestic chips in AI model inference tasks.