Literature includes several empirical models to predict shear capacity of steel fiber-reinforced concrete (SFRC) beams. However, some of these models were calibrated based on small database of experiments or include data for beams failed in flexure. This study utilizes a database of 241 tests of SFRC beams failed in shear along with gene expression programming (GEP) to establish more accurate and reliable predictions. Independent variables included in the GEP were fiber factor, longitudinal reinforcement ratio, concrete cylinder compressive strength, and shear span-to-depth ratio. The model provides an average ratio (vtest/vpredicted) of 1.00, a coefficient of variation of 24%, and a root mean square error of 1.33 MPa. The proposed model provides a higher accuracy when compared to existing models in the literature. The study also presents a reliability analysis study to evaluate the safety level embedded in the proposed model. By comparing the safety level of the proposed model to the safety level in reinforced concrete beams without shear reinforcement according to ACI 318-19 equation, a strength reduction factor of 0.5 instead of 0.75 is proposed to achieve consistent safety level with conventional reinforced concrete. Additionally, a varying strength reduction factor with respect to shear span-to-depth is also proposed.