During Huawei Connect 2025, Huawei, in collaboration with Zhejiang University, proudly presented the nation's inaugural DeepSeek-R1-Safe foundational large model, which is built upon the Ascend Thousand-Card Computing Power Platform. Ren Kui, a scholar from Zhejiang University, elaborated on the pivotal innovations embodied by the DeepSeek-R1-Safe-671B model. The dedicated research and development team has crafted a comprehensive, secure post-training framework. This marks the first instance of successfully conducting secure training for a full-scale large model with hundreds of billions of parameters on the Ascend Thousand-Card Computing Power Platform. Tests have revealed that the model boasts an impressive 83% comprehensive security defense capability, outperforming numerous contemporary counterparts. Moreover, it maintains performance degradation within a mere 1% in general capability benchmark tests, striking an optimal balance between security features and overall performance. Chen Chun, an academician of the Chinese Academy of Engineering, emphasized that the launch is intended to foster the synchronized advancement of large model security and industrial ecosystems, expressing his aspirations for more profound collaboration between the team and Huawei. Zhang Dixuan, President of Huawei's Ascend Computing Business, unveiled the open-source and open strategy for Ascend's new generation of AI hardware and comprehensive software suite. Presently, the model has been made fully open-source across multiple communities.