This study compares existing concrete permeability models across both diffuse and localized cracking regimes, evaluating their validity and applicability. Analysis of models utilizing the damage variable, D, for permeability evolution and those employing crack opening calculations is performed. Comparative analysis using splitting test measurements highlights the limitations of current models in accurately predicting permeability across varied cracking scenarios. A refined permeability model, derived from existing frameworks, is proposed to address these limitations. Measurements based on splitting tests will be used for the comparative study and the validation of the proposed model. The mechanical description of this splitting test is performed using Mazars damage model and two different regularization techniques. The efficiency of the two techniques is investigated. Moreover, various methods to estimate concrete permeability are examined and statistically analysed. Results reveal that the original non-local damage model gives spurious diffused damage and strain profiles at failure. This led to significant overestimation of transport properties. In contrast, the stress based non-local model provides localized damage and strain profiles at failure (for damage equal to 1). This is more convenient to compute a crack opening and to accurately estimate permeability. With an elasto-damageable model, numerical damage parameter is overestimated and consequently, the permeability models that are only damage based tend to overestimate permeability. Finally, the proposed model estimated permeability of concrete in a splitting setup with a R^2 of 0.999 and RMSE of 7.28×10^(-16) compared to experimental data.