July 31, 2026 – Huawei has officially announced the full open-source release of its 505-billion-parameter openPangu-2.0-Pro large language model, making the complete model weights, foundational inference code, and accompanying technical report publicly accessible across mainstream open-source platforms for industry-wide adoption.
As Huawei’s flagship open AI model brand built entirely on Ascend-native training and inference architectures, openPangu is designed to deliver production-ready best practices that lower the barrier for developers to fully leverage Ascend computing power. The initiative aims to establish a robust, scalable intelligent infrastructure tailored for the upcoming Agent era, accelerating commercial AI innovation and fostering a thriving, collaborative open-source ecosystem.
All components of the openPangu-2.0 series have now been rolled out across leading global code repositories, granting unrestricted free access to every developer, enterprise partner, and academic researcher worldwide. Users can directly download the full model package for local deployment, or experience the model’s capabilities in real time via Huawei Cloud’s MaaS (Model as a Service) platform. Huawei also encourages the global community to submit feedback on real-world usage scenarios, driving continuous iterative improvements to the open-source ecosystem.
During this June’s Huawei Developer Conference, Richard Yu, Huawei’s Executive Director, walked through the full R&D journey of the Pangu large model family. He emphasized that Huawei launched China’s very first domestically developed large model at a time when most industry practitioners globally were still working to grasp the core concepts of large-scale AI systems, cementing the company’s position as an early pioneer in the global large model race.
