Multi-Agent Deep Reinforcement Learning (MADRL)-based scheduling has become a promising framework for optimizing decision-making processes in complex systems, such as 5G wireless networks. While RL excels in environments with small state and action spaces, scalability challenges arise in large-scale scenarios due to difficulties in accurately estimating state and action values. This paper addressed these challenges by proposing a hierarchical RL-based approach tailored for task scheduling in 5 Generation Cloud RAN, integrating Edge, Task, Workload, and Energy Efficiency scheduling. A multi-agent deep reinforcement learning (MADRL) framework was proposed for task scheduling in 5 Generation Cloud RAN. The coordinator handles task-to-server assignment, while each Server Agent manages local CPU allocation. The system is integrated with an NS-3 5G LENA simulation to evaluate latency, energy consumption, throughput, and queue stability under varying user loads. Experiments using 20 random seeds and confidence intervals showed that the proposed MADRL architecture achieved better latency–energy trade-offs and improved scalability compared to baseline schedulers. The hybrid approach demonstrated significant potential for balancing global coordination with localized decision-making, offering a robust solution for dynamic resource optimization in next-generation wireless networks.