Nuclear data uncertainty analysis on the spent nuclear fuel inventory was performed on the Takahama-3 NT3G23 assembly, where the sample SF95-4 was irradiated up to a burnup of approximately 36 GWdt according to the SFCOMPO benchmark. The cross-section covariance matrices stored in the ENDF/B-VIII.0, JEFF-3.3 and JENDL-4.0u evaluated nuclear data libraries were propagated with the stochastic sampling algorithms implemented in the SANDY code. A comparison of the concentration uncertainty differences obtained using data from the three libraries is reported. Similarities were found with the fuel composition uncertainty results obtained for the Calvert Cliffs MKP109 sample P SFCOMPO benchmark. Such a similarity was also found when comparing concentration uncertainties along the sample irradiation. Therefore, the main contributors to the concentration uncertainty of a number of nuclides were identified at different burnup levels in the two samples. To complement the similarity analysis, a correlation study of the concentration distributions predicted by the two models was performed. The reported results hint a dominance of the common uncertainty propagation mechanisms over the model differences in the determination of concentration uncertainty.