Headphone-based reproduction of Ambisonic sound scenes, also known as binaural Ambisonic rendering, provides an effective method for real-time renderings, such as rendering dynamic sound scenes caused by head rotation. However, the low-order Ambisonic decoding results in the degradation of the rendering quality, while the high-order decoding increases the computation complexity. A mixed-order Ambisonic (MOA) rendering scheme is proposed in this paper, which uses only the most relevant direction-related terms for sound field rendering. The proposed method is compared with traditional higher-order Ambisonic (HOA) rendering in terms of modeling error and computational complexity. To evaluate auditory perception, four participants are invited to participate in a listening experiment. The results demonstrate that using a direction-related MOA scheme can reduce computation complexity for dynamic binaural rendering without affecting auditory perception.