We study the redshift distribution of two samples of early-type gravitational lenses, extracted from a larger collection of 122 systems, to constrain the cosmological constant in the {Lambda}CDM model and the parameters of a set of alternative dark energy models (XCDM, Dvali-Gabadadze-Porrati, and Ricci dark energy models), in a spatially flat universe. The likelihood is maximized for {Omega}_{Lambda}_=0.70+/-0.09 when considering the sample excluding the Sloan Lens ACS systems (known to be biased toward large image-separation lenses) and no-evolution, and {Omega}_{Lambda}_=0.81+/-0.05 when limiting to gravitational lenses with image separation {Delta}{theta}>2" and no-evolution. In both cases, results accounting for galaxy evolution are consistent within 1{sigma}. The present test supports the accelerated expansion, by excluding the null hypothesis (i.e., {Omega}_{Lambda}_=0) at more than 4{sigma}, regardless of the chosen sample and assumptions on the galaxy evolution. A comparison between competitive world models is performed by means of the Bayesian information criterion. This shows that the simplest cosmological constant model--that has only one free parameter--is still preferred by the available data on the redshift distribution of gravitational lenses. We perform an analysis of the possible systematic effects, finding that the systematic errors due to sample incompleteness, galaxy evolution, and model uncertainties approximately equal the statistical errors, with present-day data. We find that the largest sources of systemic errors are the dynamical normalization and the high-velocity cutoff factor, followed by the faint-end slope of the velocity dispersion function.