In order to improve the cleanliness of steel, non-aluminum deoxidation processes have begun to replace aluminum deoxidation processes. Although the aluminum deoxidation process can reduce the oxygen content in steel to <10 × 10−6, this deoxidation method causes fatigue failure resulting from the formation of large-grained spherical (Ds-type) inclusions composed of calcium–aluminate. It also tends to lead to nozzle blockage during casting. Given the above problems, this study conducted an in-depth investigation of silicon–manganese deoxidation. Thermal experiments and thermodynamic calculations were used to assess the impact of different Mn–Si ratios on the oxygen content and inclusion characteristics during the deoxidation process of molten steel with different initial oxygen contents. The experimental samples were analyzed using an oxygen–nitrogen–hydrogen analyzer, a direct reading spectrometer, and an automatic scanning electron microscope. After that, the samples were electrolyzed to observe the 2D morphology and 3D morphology of the inclusions using scanning electron microscopy. Finally, thermodynamic calculations were carried out using FactSage to verify the experimental results. The results indicated that, regardless of the initial oxygen content, silicon–manganese deoxidation maintained the total oxygen content at 35 × 10−6. It effectively managed the plasticization of inclusions in molten steel, predominantly yielding spherical silicates while minimizing Al-containing inclusions. Nevertheless, as the initial content of [O] increased, the size and density of the silicate inclusions in the steel also increased. An optimal point in the number and size of inclusions was observed with an increased Mn–Si ratio. Moreover, the combined utilization of silicon–manganese deoxidation, diffusion deoxidation, and vacuum deoxidation enabled ultra-low oxygen content control of molten steel.