Huawei Cloud has released Pangu Models 5.5, a 718-billion parameter deep thinking model designed specifically for industrial applications across manufacturing, scientific research, and agriculture.

Huawei Cloud CEO and executive director of Huawei Zhang Ping’an announced the new models at the Huawei Developer Conference 2025, highlighting their implementation across more than 500 scenarios in more than 30 industries.

At the conference, the company also announced a brand upgrade and unveiled the next-generation Huawei Cloud AI Cloud Service, powered by CloudMatrix 384 supernodes to enhance compute capabilities for advanced AI applications.

The Pangu 5.5 models have been upgraded with five basic models including natural language processing, computer vision, multi-modal, prediction, and scientific computing.

Pangu Models offer six core capabilities to enterprises, allowing them to build professional models without starting from scratch. The Chinese Academy of Agricultural Sciences has utilised Pangu foundation models to develop an Agricultural Scientific Discovery Model, which has significantly improved crop breeding processes.

The Pangu World Model, based on the Pangu Multimodal Model, creates digital spaces for training AI robots and intelligent driving, supporting continuous optimisation.

In partnership with Guangzhou Automobile Group, Huawei Cloud has achieved pixel-level mapping for intelligent driving training, expediting end-to-end model iteration. The CloudRobo Embodied AI Platform, leveraging Pangu Models, offers end-to-end capabilities for embodied AI innovation and proposes the Robot to Cloud (R2C) Protocol to foster an open ecosystem for robotics development.

The Pangu Prediction Model, with its unique triplet transformer architecture, improves prediction accuracy and generalisation across industries. This model has been successfully applied by companies like Conch Cement and China Baowu Steel Group to optimise processes and reduce costs.

The Pangu Scientific Computing Model has enhanced weather forecasting and energy yield prediction, while the Pangu CV Model, the largest CV model in the industry, supports multi-dimensional perception and decision-making.

Huawei Cloud’s brand upgrade reflects its commitment to AI innovation and industry transformation.

The event also introduced ModelArts Versatile, an AI agent platform, and Pangu Doer, an intelligent assistant that enhances customer service and development efficiency.

The new Huawei Cloud AI Cloud Service is designed to meet the growing compute demands of AI foundation model training and inference. The service’s supernode architecture, a first in the industry, allows for peer-to-peer interconnection of proprietary NPUs and Kunpeng CPUs, boosting inference throughput significantly.

This architecture supports mixture of experts models, with one expert per card, enabling concurrent inference of 384 experts. It also optimises resource allocation and model FLOPS utilisation, enhancing efficiency by over 50%.

For large-scale model training tasks, the supernode system can form an ultra-large cluster with up to 160,000 cards, providing integrated compute for both training and inference. This flexible allocation helps customers optimise resource usage.

Huawei Cloud AI Cloud Service now serves more than 1,300 customers, including Sina and the Chinese Academy of Sciences, facilitating intelligent industry upgrades.