Deploying a 5G network in a city used to mean a team of people conducting on-site inspections, manually collecting data, and relying on experts to select sites one by one. This process was inefficient, time-consuming, and difficult to achieve the optimal overall solution. On July 21, according to a report from China Telecommunications Daily, China Telecom has handed this arduous task to a large model—its 5G wireless network planning large model has completed field trials, improving planning efficiency by 50% and achieving an accuracy rate of over 75% for the planning schemes, realizing automatic output of design plans and accurate prediction of construction results.

Traditional wireless network planning heavily relied on human experience and physical effort. In recent years, the industry has generally turned to big data analysis and ray tracing model simulations to speed up the process. China Telecom has taken another step forward in this regard. It responds to the Ministry of Industry and Information Technology's requirements for the innovation and development of "AI plus information and communication", thoroughly integrating the entire process from network insight, site planning, twin simulation, to scheme generation, building a set of digital employee capabilities covering full-scenario perception, analysis, decision-making, and execution. This focuses on three key aspects: multi-modal data integration, atomic capability intelligent orchestration, and precise simulation of dynamic business changes.

The foundation of this capability first resulted in a high-quality network planning dataset with multiple features. China Telecom integrated multi-source heterogeneous data such as performance, configuration, geography, population, and service experience, using knowledge graphs, rule engines, NLP information extraction, and LLM prompt engineering technologies to systematically translate the experience of planning experts into structured data assets. Then, it used smart data parsing tools to ensure data quality, laying a solid foundation for intelligent planning.

More importantly, there is a pioneering architecture—double-twin large model collaborative planning. The channel twin model is responsible for high-precision reconstruction of the network space and forward-looking site planning, while the traffic twin model accurately predicts traffic fluctuations in complex scenarios. These two models work together to optimize and automatically generate the best plan that takes into account network coverage, capacity, value, and structure. This is the first time that network planning has truly shifted from experience-based to model-driven.

Having just the architecture isn't enough. China Telecom conducted real-world testing with partners. It verified the technology in complex scenarios such as residential areas and underground parking lots in Pudong, Shanghai, with ZTE Communications. In residential areas, the model integrated multi-dimensional data such as population density, building distribution, and grid MR, achieving a station placement accuracy of over 80%. In underground parking lots, the team innovatively introduced the correlation between mobile speed and signal attenuation characteristics, achieving a coverage issue identification accuracy of over 75%.

Notably, there is also a strategic shift. At the 2026 China Internet Conference this month, China Telecom's vice president Luan Xiaowei stated in his speech that China Telecom will actively embrace and fully integrate into intelligence, continuously improve the integrated system of computing power, platform, data, large models, and intelligent entities, and promote the transformation of enterprise operations from traditional traffic operations to Token operations. When the site selection of base stations begins to be decided by large models, the operating logic of telecom operators is also being quietly rewritten.