In response to the limited adaptability of conventional evasive tactics, a Soft Actor-Critic (SAC) based penetration strategy for high-speed vehicles is proposed. The strategy is tailored for a three-dimensional (3D) combat scenario where two interceptors commence with stochastic initial postures. The reward function is designed to prioritize successful penetration while concurrently minimizing energy consumption and trajectory deviation during evasive maneuvers. The algorithm enables the high-speed vehicle to make online decisions based on real-time relative motions and the vehicle’s intrinsic status, issuing optimal overload commands for evasion. The guidance laws for altitude and heading tracking to return to the original trajectory are designed. The simulation results indicate that the proposed method is applicable to frontal encirclement interception scenarios under various initial conditions, achieving a success evasion rate of over 90% while adhering to energy constraints. Additionally, the method ensures that the vehicle maintains minimal deviation from its original trajectory post-evasion, facilitating a swift return to the initial flight path.