In Vehicular Internet of Things (VIoT) environments, vehicles with limited computational resources often need to offload tasks to other vehicles or edge servers with surplus capacity. However, the highly dynamic and mobile nature of VIoT networks poses significant challenges to guaranteeing timely and efficient task offloading. To address this, we propose a novel approach called Vehicular Internet of Things Task Offloading (VIoT-TO), which partitions the network into a cellular structure and employs reinforcement learning to determine optimal task offloading strategies. Specifically, the system learns how to identify nearby idle task servers either peer vehicles or edge servers within each network cell. The proposed method leverages Q-learning to solve the reinforcement learning problem, with the task offloading modeled as a Markov Decision Process (MDP). The reward function is carefully designed to encourage fair load distribution across servers while also prioritizing servers that are geographically closer, thereby reducing communication latency. As a result, the overall task offloading delay becomes more predictable and manageable. Experimental evaluations demonstrate that VIoT-TO outperforms existing benchmark approaches in terms of task offloading delay, load balancing efficiency, and task completion rate. These findings suggest that VIoT-TO is an effective and scalable solution for real-time task management in vehicular networks.