With the advent of the 6G era, communication systems are encountering major challenges, including the explosive growth in data scale and complexity as well as the rapidly increasing demand for real-time intelligent decision-making. Traditional communication paradigms based on bit transmission and discriminative artificial intelligence (AI) models struggle to meet the future network’s requirements for generalization, robustness, and adaptability. To achieve a paradigm shift from “connected intelligence” to “cognitive intelligence”, Agentic AI, which is powered by Large AI Models (LAMs), is emerging as a key direction for driving the intelligent evolution of 6G networks.

LAMs possess powerful capabilities for knowledge abstraction, cross-modal understanding, and content generation, enabling them to efficiently perform critical tasks in 6G networks such as semantic communication, resource scheduling, channel modeling, and network optimization. However, conventional LAMs typically rely on offline training and lack the abilities of real-time sensing, continual learning, and autonomous optimization, making them ill-suited for dynamic and complex communication environments. Against this backdrop, the concept of Agentic AI has emerged. The core idea of Agentic AI is to use large models as the “cognitive core”, while integrating key modules for perception, memory, decision-making, and planning to construct intelligent agent systems with environmental interaction, autonomous reasoning, and dynamic optimization capabilities. By forming a closed loop of “perception–cognition–action–feedback”, Agentic AI enables adaptive decision-making and efficient collaboration across multi-task, multi-modal, and multi-scenario 6G networks.

Although the potential of LAMs and Agentic AI in 6G communication systems is vast, their practical deployment faces several critical challenges. These include achieving efficient collaboration and control among multiple intelligent agents, enabling robust task planning and decomposition in dynamic environments, developing unified frameworks for agent performance evaluation, and ensuring data security and trustworthy privacy protection. This Special Issue (SI) aims to systematically investigate the key scientific and engineering challenges associated with integrating LAMs and Agentic AI into 6G networks. The focus is on uncovering the underlying integration mechanisms, establishing solid theoretical foundations, and exploring the application prospects of these technologies within communication systems. By doing so, the project seeks to advance both fundamental understanding and technological innovation in intelligent 6G networks. Topics of interest include, but are not limited to

•  LAMs and Agentic AI for Semantic Communication and Intelligent Perception
•  LAMs and Agentic AI for Multi-Agent Communication Coordination and Resource Management
•  LAMs and Agentic AI for Adaptive Network Slicing and System-Level Optimization
•  LAMs and Agentic AI for Digital Twins and Cyber–Physical Integrated Communications
•  Multimodal Fusion for Collaborative Agent Mechanisms in Intelligent Communication Systems
•  Memory- and Knowledge-Augmented Approaches for Communication Task Planning and Decision Reasoning
•  LAMs and Agentic AI for Continual Learning and Autonomous Optimization in 6G
•  LAMs and Agentic AI for Channel Modeling and Network Design
•  LAMs and Agentic AI for 6G Network Security, Privacy, and Trustworthy Communications
•  LAMs and Agentic AI for Communication Prototype Systems and Validation Platforms

Submission Guidelines

Authors should prepare papers in accordance with the format requirements of Tsinghua Science and Technology, with reference to the Instruction given at  https://www.sciopen.com/journal/1007-0214, and submit the complete manuscript through the online manuscript submission system at https://mc03.manuscriptcentral.com/tst with manuscript type as “Special Issue on From Large AI Models to Agentic AI: A Road Map to 6G”.

Important Dates

Deadline for submissions: March 31, 2026

Guest Editors

Cunhua Pan, Southeast University, China

Feibo Jiang, Southeast University, China

Kezhi Wang, Brunel University of London, U.K.

Marco DI RENZO, CentraleSupelec, France

Dusit Niyato, Nanyang Technological University (NTU), Singapore

Ekram Hossain, University of Manitoba, Canada