With the continuous breakthroughs of large models in multimodal perception, complex reasoning, and autonomous decision-making, 6G networks are rapidly evolving toward a new paradigm characterized by native intelligence, task-driven operation, and autonomous evolution. This report systematically reviews the integration pathways of large models and agent-based technologies in future 6G networks, with a particular focus on emerging paradigms such as vision–language–action (VLA) models, world models, as well as key agentic technologies including Agentic RAG, harness engineering, and the Internet of Agents. Building upon this foundation, the report highlights two representative directions for empowering 6G networks with large models and intelligent agents. First, it introduces the concept of large model-driven token communication, where token-level semantic representation and transmission enable a unified modeling framework for multimodal and multitask communication systems. Second, it explores agentic systems for communication networks, which support autonomous task planning and dynamic resource orchestration, thereby achieving a fine-grained balance among intelligence capability, resource consumption, and latency. Finally, looking ahead to future 6G networks, the report analyzes the key challenges and emerging opportunities in constructing autonomous, efficient, trustworthy, and evolvable intelligent communication systems.