Recently, on the day of the official release of the DeepSeek V4 large model, CHALCO took the lead in deploying the DeepSeek V4 Flash version in the national production environment with the help of its independently developed Kun'an AI platform, and simultaneously integrated it into the Xiaokun AI assistant and Xiaokun intelligent agent platform, achieving zero time difference from model access to business empowerment. This is a crucial step towards building an independent, controllable, secure, and efficient AI intelligent computing base for central enterprises.

Artificial intelligence accelerates integration into the real economy

Currently, artificial intelligence is accelerating its integration into the core areas of the real economy, and the central enterprise AI+special action continues to deepen. As a leading enterprise in the non-ferrous metal industry, CHALCO's business covers the entire industry chain of mineral resource development, material research and development manufacturing, green and low-carbon transformation, etc. It has extremely high standards for domestic adaptation of large model applications, data security compliance, and industrial scenario adaptation. Faced with the three major challenges of stable operation of core models in a full stack domestic environment, high-density computing power support for ultra large scale parameter models, and efficient context processing for industrial long texts, CHALCO uses Kun'an AI middleware as the core carrier to systematically promote the integration and implementation of national cloud+national chip+national model.

Empower all business scenarios

The entire deployment process relies on the national computing power cluster and the national industrial software environment. The Huawei Xiaoxiaoling team provides low-level computing power optimization and model migration technical support, while China Mobile assists in network optimization and hardware expansion. The three parties work together to overcome key technical difficulties such as domestic chip adaptation and distributed inference acceleration. Among them, the DeepSeek V4 Flash version, with its lightweight architecture and efficient reasoning capabilities, achieved seamless integration with the Xiaokun AI assistant on the day of its release, providing real-time services such as intelligent Q&A, knowledge retrieval, and process assistance for frontline employees; At the same time, the upgraded Xiaokun intelligent agent platform supports business departments to quickly build customized AI applications, promoting the deep penetration of big model capabilities into various aspects such as production, management, and research and development.

Strengthen decision support

The DeepSeek V4 Pro version, which will be deployed on May 6, 2026, further enhances CHALCO's intelligent decision support capabilities in complex scenarios. This version relies on the 1.6 trillion parameter MoE architecture to achieve high concurrency and low latency stable inference in a localized environment. It natively supports ultra long context processing and can efficiently parse professional materials such as geological exploration reports, process documents, equipment operation manuals, etc., providing strong support for technical breakthroughs, risk prediction, and knowledge accumulation. At present, two versions of the model have achieved unified management and usage statistics in Kun'an AI platform, balancing resource utilization efficiency and business response speed. Keywords: DeepSeek, Chinalco Group, Digital Transformation

The rapid implementation of the DeepSeek V4 series models is an important practice for CHALCO to implement national artificial intelligence+action and promote digital transformation. Next, Chinalco Group will continue to deepen the construction of Kunan AI platform capabilities, improve the center edge collaborative architecture, expand core functions such as multimodal data processing, codeless training, and cloud edge collaboration, accelerate the construction of a central enterprise level AI infrastructure covering the entire lifecycle of models, computing power full stack scheduling, and security domain control, and inject stronger intelligent power into the high-quality development of the non-ferrous metal industry.Editor/Gong Ziwei