[Huatai Securities: DeepSeek-V4 Lowers Long-Context Costs, Driving Demand Expansion] Huatai Securities' research report states that on April 24, DeepSeek released its V4 series model and accompanying paper. The core change focuses on the low-cost practical implementation of 1M token contexts, with systematic upgrades to model architecture, training processes, and infrastructure. Huatai Securities believes the market may easily interpret V4 as 'reducing costs and lowering computational and storage demands,' but the more significant marginal change lies in the reduced cost of long-context processing. This enhances the usability of scenarios such as complex agents, multi-document analysis, long-term tasks, and online learning, potentially driving an increase in inference calls and storage access frequency. Potential catalysts include Day 0 adaptation for domestic computing power, subsequent scaling of the Ascend 950 supernodes, advancements in DeepSeek's multimodal and Engram roadmap, and the industrial development driven by interoperability between domestic open-source model algorithms. (People's Financial News)