In this Letter, we present an analysis of the performances of a local density functional, recently proposed by Ragot and Cortona. This functional, derived using the Colle-Salvetti approach, is parameter free, not containing any fitted parameters. The results obtained coupling this functional with the traditional Slater exchange, show that it outperforms the most common local approximations and significantly reduces the performance gap between local and gradient-corrected functional. This local approximation has been also cast in a hybrid functional, leading to the mB3LYP model. This approach works at least as well as the B3LYP functional, but it is based on a more coherent physical model, since both gradient corrected and local ingredients stem from the Colle-Salvetti model. (C) 2007 Elsevier B.V. All rights reserved.