Thermal transport in magnetic materials has attracted tremendous attention due to its importance in spintronics and thermal management. However, existing calculations of thermal conductivity of magnetic materials often simplify the complex non-collinear spin structures and realistic magnetic moment orientations by assuming collinear spin structures with spins aligned along the z-axis. To evaluate the reliability of such simplification, by means of machine learning potentials (MLPs) combined with the Boltzmann transport equation (BTE), we systematically investigate the thermal transport properties of monolayer BiOsO3, a ferromagnetic semiconductor with strong spin–orbit coupling (SOC) and large magnetic anisotropy (MA). The calculations show that SOC effect could significantly enhance the lattice thermal conductivity (2.83 W/mK) by a factor of approximately 2.2 compared to the case without SOC (1.29 W/mK). This obvious enhancement mainly originates from the fact that the introduction of SOC drives charge redistribution toward the inner Os–O bond network, thereby enhancing the symmetry of internal potential wells and suppressing phonon anharmonicity. Such behavior indicates that the previous simplified collinear approaches are insufficient to accurately describe the thermal transport in magnetic materials with strong SOC. In contrast, BiOsO3 exhibits a large magnetic anisotropy energy (MAE=−5.7 meV/Os). However, the orientation of spin easy axis hosts a weak influence on the thermal conductivity. These findings shed light on the thermal conductivity of BiOsO3, and could also provide meaningful guidance for studying thermal transport in magnetic materials.