Unveiled by Zhang Ping’an, executive director of Huawei, CEO of Huawei Cloud, the update includes enhancements across five core capabilities: natural language processing (NLP), computer vision (CV), multi-modal processing, prediction, and scientific computing.
“Pangu Models help customers tackle the most challenging issues in their specific scenarios and reimagine both operations and efficiency across numerous industries,” said Zhang.
Among the highlights is the Pangu NLP Model, now featuring a powerful 718B MoE (Mixture of Experts) architecture. It offers advanced tool invocation, reasoning, and maths capabilities, while improving long-sequence processing and minimising hallucinations.
The DeepDiver module can handle over 10-step Q&A tasks within five minutes and generate detailed reports exceeding 10,000 words.
Huawei is also emphasising customisability. Pangu Models enable businesses to build proprietary AI systems without “reinventing the wheel”, providing core capabilities such as pre/post-training corpora, engineering tools, judge models, and industry evaluation platforms. With ModelArts, companies can tailor their own professional models using internal data.
The models are already in action. The Chinese Academy of Agricultural Sciences (CAAS) used Pangu to develop an agricultural research model that cut crop R&D times and improved rice strains with reduced height and better lodging resistance.
CAAS reported that this resulted in the plant being approximately 25% shorter than conventional rice varieties, while greatly enhancing its resistance to lodging.
At the event, Zhang introduced five new deep-thinking models for the medical, finance, government, industrial, and automotive sectors, scheduled for release in June.
The Pangu World Model, based on Huawei’s multimodal technology, enables simulation of driving environments using video and lidar point clouds. In partnership with GAC Group, it delivers pixel-level mapping between 2D and 3D data to support rapid iteration of intelligent driving systems.
Meanwhile, the Pangu Prediction Model employs a triplet transformer pre-training framework for handling diverse data types, improving cross-industry prediction accuracy. Conch Cement uses it to optimise raw material use, allowing sustainable reuse of construction waste.
For scientific computing, Huawei has extended Pangu’s capabilities in meteorology and renewable energy. Shenzhen’s meteorological bureau and Chongqing’s Tianzi model now benefit from more accurate extreme weather forecasts. Shenzhen Energy Group uses the model for better solar and wind output predictions.
The new 30B-parameter Pangu CV Model supports multi-modal perception and creates rare-case fault sample libraries. CNPC’s Kunlun model, built on Pangu, is already deployed in over 100 domains, detecting oil pipeline defects with sub-millimetre precision and reducing manual labour by 25%.
To date, Pangu Models have been applied in over 500 use cases across 30 industries, marking a significant step in accelerating intelligent transformation across sectors worldwide.
RELATED STORIES
How Huawei Cloud Stack is shaping the future of AI & hybrid cloud
Huawei, TM Forum unveil White Paper 3.0, marking a major leap in intelligent network operations
Huawei expands AI cloud services in Asia-Pacific despite US sanctions
