<p>To address the challenge of co-channel interference in computation-intensive task offloading for vehicular edge computing (VEC), this paper proposes a novel task offloading and resource allocation algorithm based on interference-aware Soft Actor-Critic (IA-SAC). First, an interference aware (IA) module is designed to monitor the real-time signal-to-interference-plus-noise ratio (SINR) and load factor of each channel. This information is then integrated with the original state to construct a multi-dimensional decision space. On this basis, a multi-branch hybrid Actor network architecture is developed to enable coordinated sampling from a hybrid action space, which includes continuous power allocation, discrete channel and offloading mode selection. This structure overcomes the limitations of conventional reinforcement learning algorithms confined to a single action space. Ablation experimental results demonstrate the IA-SAC algorithm achieves a 7.72% reduction in task processing delay compared to the baseline SAC algorithm. In comparative experiments of reinforcement learning algorithms, the IA-SAC algorithm reduces the total task processing delay by 12.42%, 11.89%, 27.23%, 34.46% and 37.72% compared to DDPG, TD3, PPO, Greedy and Dueling DDQN, respectively. Furthermore, these findings demonstrate that the proposed IA-SAC based task offloading algorithm effectively reduces the total task processing delay and improves resource utilization.</p>