Earthquake recurrence models are the basis of seismic hazard analysis and seismic risk evaluation of physical infrastructure. They are based on statistical analysis of earthquake occurrence data available in a specific geographical region. This work proposes a new earthquake recurrence model for the interface and intraslab seismicity of the subduction margin along Chile. The model improves some of the shortcomings of previous available models in the region such as the lack of earthquake declustering or the use of magnitude scales inconsistent with modern ground‐motion prediction equations. Significant differences in seismic rates are found with some previous models. Indeed, the resulting frequencies from the Gutenberg–Richter relations are not only similar to some of the previous works, but also one order of magnitude higher and lower than two of the previously reported models. Because one of the strongest assumptions in earthquake occurrence models is that they follow a homogeneous Poisson process, this hypothesis is statistically tested herein, finding that the declustered catalog only partially complies with this assumption, showing for instance that the interevent times follow approximately an exponential distribution.