Categories: Technology

Huawei Cloud rolls out Pangu Models 5.5 to cover more industries

Huawei Cloud has rolled out Pangu Models 5.5 at the Huawei Developer Conference 2025, introducing a new round of updates focused on helping businesses solve real-world problems with AI. The latest version includes upgrades in five key areas: natural language processing (NLP), computer vision (CV), multimodal modelling, prediction, and scientific computing.

Zhang Ping’an, executive director of Huawei and CEO of Huawei Cloud, said the Pangu lineup is built with a goal of making AI more useful for companies in specific sectors. The focus is less on general-purpose tools and more on building models that can perform well with real-world data in complex environments like farms, factories, and city infrastructure.

A closer look at NLP improvements

The new Pangu NLP model features a 718-billion-parameter MoE (mixture of experts) design. It has 256 experts that the system can call on depending on the task. According to Huawei, it’s better at reasoning, tool use, and mathematical tasks.

The model is built to work with longer sequences of text, reduce hallucinations, and switch between quick replies and deeper thinking as needed. The adaptive setup allows it to give faster answers to simple questions while taking more time on harder ones – an approach the company says can boost overall efficiency by eight times.

Huawei has also introduced DeepDiver, a tool that supports long, multi-step question answering and information gathering. DeepDiver can complete tasks with more than 10 steps in under five minutes and generate detailed reports of over 10,000 words. The idea is to reduce the amount of manual effort needed for knowledge work, particularly in roles that rely on research and document creation.

Tools for building custom models

To support enterprise use, Huawei Cloud offers six core elements for building industry models. These include foundation and industry-specific models, curated datasets, data engineering & training tools, and benchmarking platforms. With ModelArts, businesses can train and refine their models using their internal data.

One real-world example comes from the Chinese Academy of Agricultural Sciences (CAAS), which used Huawei’s models to build its own AI system for agricultural research. The model can answer questions based on scientific litreature and genetic data, run gene analysis, and design experiments. Using this approach, CAAS scientists improved a rice variety that’s 25% shorter than conventional strains, making it more resistant to wind damage without reducing crop yield.

Domain-specific models

Huawei is also launching models tailored to five industries – healthcare, finance, government, industrial manufacturing, and automotive. These are expected to go live in June. Each model is trained with domain-specific data and tuned for tasks that are common in that sector.

In automotive, for example, the new Pangu World Model generates synthetic driving data from maps, sensor input, and vehicle commands. This creates a low-cost alternative to real-world testing. Guangzhou Automobile Group (GAC Group) is using the model to simulate lidar and video data, allowing developers to recreate rare driving scenarios and test software faster.

Expanded use in prediction and scientific computing

The Pangu Prediction Model uses a triplet transformer architecture to process tables, logs, and images in the same framework. Companies like Conch Cement are using this to predict product strength and optimise raw material mixes. The system supports reusing waste like demolished concrete and industrial by-products while maintaining output quality.

In weather forecasting, local meteorology bureaus in Shenzhen and Chongqing have adopted Pangu to improve short-term rain prediction. Energy companies, including Shenzhen Energy, are using the same platform to forecast wind and solar power output, helping them balance energy loads more effectively.

Upgrades in computer vision

Huawei has also released a 30-billion-parameter CV model using a new MoE structure. The model supports multiple input types, including images, infrared, lidar, radar, and spectrum data. It’s built to detect problems in industrial settings – like small cracks or heat anomalies in equipment – that are hard to spot with the human eye.

CNPC, China’s national oil company, has applied the model in areas like oilfield development, chemical production, and equipment maintenance. In manufacturing, the system detects faults with sub-millimetre precision, boosting detection rates by 40% and cutting manual effort by around 25%.

Want to learn more about cybersecurity and the cloud from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is co-located with other leading events including Digital Transformation Week, IoT Tech Expo, Blockchain Expo, and AI & Big Data Expo.

Explore other upcoming enterprise technology events and webinars powered by TechForge here.

Tags: ai, developer, development, huawei