We present the results of an analysis aimed at testing the accuracy and precision of the PARSEC v1.2S library of stellar evolution models, combined with a Bayesian approach, to infer stellar parameters. We mainly employ the online DEBCat catalogue by Southworth, a compilation of detached eclipsing binary systems with published measurements of masses and radii to ˜2 per cent precision. We select a sample of 318 binary components, with masses between 0.10 and 14.5 M⊙, at distances between 1.3 pc and ˜8 kpc for Galactic objects and ˜ 44-68 kpc for the extragalactic ones. The Bayesian analysis applied takes as input effective temperature, radius, and [Fe/H], and their uncertainties, returning theoretical predictions for other stellar parameters. From the comparison with dynamical masses, we conclude that the inferred masses are precisely derived for stars on the main-sequence and in the core-helium-burning phase, with uncertainties of 4 per cent and 7 per cent, respectively, on average. Masses for the subgiants and red giants are predicted within 14 per cent, and those for early asymptotic giant branch stars within 24 per cent. These results are helpful to further improve the models, in particular for advanced evolutionary stages for which our understanding is limited. We obtain distances and ages for the binary systems and compare them, whenever possible, with precise literature estimates, finding excellent agreement. We discuss evolutionary effects and challenges for inferring stellar ages from evolutionary models. We also provide useful polynomial fittings to theoretical zero-age main-sequence relations.