Huawei Open-Sources Pangu AI Models to Boost Ascend Ecosystem
Huawei Releases Open-Source AI Models to Fuel Innovation
Chinese tech giant Huawei has taken a significant step in advancing artificial intelligence development by open-sourcing two cutting-edge large language models (LLMs) on June 30, 2025. The company made publicly available both the Pangu 7B parameter dense model and the more advanced PanguPro MoE (Mixture of Experts) 72B parameter model, along with their proprietary Ascend-based inference technology.
Strategic Move for Ecosystem Growth
This open-source initiative represents a key component of Huawei's strategy to build out its Ascend computing ecosystem. By making these powerful AI models accessible to developers worldwide, Huawei aims to:
- Accelerate research and innovation in large model technologies
- Foster cross-industry AI applications
- Create collaborative opportunities for value creation
- Strengthen the global developer community around Ascend hardware
"We believe open collaboration is essential for advancing AI capabilities," stated a Huawei spokesperson. "By sharing these resources, we hope to spark new innovations that benefit enterprises and society as a whole."
Technical Specifications and Availability
The released resources include:
- PanguPro MoE 72B model weights
- Basic inference code for standard implementations
- Specialized inference code optimized for ultra-large-scale MoE models running on Ascend hardware
The Pangu 7B model weights and corresponding inference code are scheduled for release in the coming weeks, according to company representatives.
Both models represent significant achievements in Chinese AI development:
| Model | Parameters | Architecture | Key Features |
|---|
The mixture-of-experts approach in particular allows for more efficient computation by activating only relevant portions of the neural network for each input.
Call for Global Collaboration
Huawei has extended an open invitation to:
- Developers worldwide to experiment with the models
- Enterprise partners to explore commercial applications
- Academic researchers to contribute improvements
The company has established a dedicated portal (https://gitcode.com/ascend-tribe) where interested parties can access resources and provide feedback. "We welcome all constructive input that can help refine these technologies," noted Dr. Li Wei, head of Huawei's AI research division. "This is just the beginning of what we hope will be a thriving open ecosystem around Ascend-powered AI."
The move comes as Chinese tech firms increasingly compete in the global AI landscape, with open-source strategies becoming a key differentiator for attracting developer mindshare.
Key Points
- Huawei open-sourced two advanced LLMs: Pangu 7B (dense) and PanguPro MoE 72B
- Includes Ascend-optimized inference technology for enterprise deployment
- Part of broader strategy to build Ascend computing ecosystem
- Available now on GitCode platform with more resources coming soon
- Company seeks global collaboration to improve technologies