Large audio-language models struggle to match the logical reasoning capabilities of their text-based counterparts, primarily due to limited high-quality audio reasoning data. Researchers have introduced X$^3$-OPD, a novel framework designed to bridge this gap by distilling reasoning knowledge from a powerful text-based teacher into large audio-language models via on-policy alignment1. This cross-modal approach enables the transfer of logical reasoning capabilities, potentially enhancing the performance of audio-language models. The X$^3$-OPD framework is built on the concept of on-policy distillation, which allows for more effective knowledge transfer between models. By leveraging this framework, audio-language models can improve their ability to reason and understand complex audio inputs. This development matters to practitioners because it has the potential to significantly enhance the reasoning capabilities of audio-language models, making them more effective in real-world applications.