Hidden Markov models (HMMs) are a powerful approach for discovering novel insights across a number of biological applications, from biological sequence analysis to molecular structure prediction to even phylogenetic tree construction. Recently, HMMs have become the primary framework for phylogenetic comparative biology as means for modeling lineage-specific heterogeneity in both trait evolution and diversification without vastly increasing the number of parameters. In our view, HMMs have enormous potential in divergence time applications that separate branch lengths (that are in units of raw substitution counts) into their two component parts of durations of time and substitution rates, which often exhibit lineage-specific variation. These lineage-specific rates complicate the estimation of divergence times, because we cannot assume that the substitution rate from one lineage accurately represents the substitution rates in other lineages in the tree. Many solutions exist to relax these assumptions, but are largely borne out of statistical convenience. More importantly, they violate our natural intuitions about how rates likely evolve across a tree just like any trait, especially if these rates are co-evolving with changes in life-history and/or environmental interactions. Here we describe an HMM modeling framework for examining not only substitution rates that follow the evolution of an observed focal trait of interest, but also the evolution of “hidden” states, allowing the discovery of correlations with other observed and unobserved factors. We will also show how our new scalable approach can discover underlying variation of evolutionary rates, facilitating more confidence and accuracy in divergence time estimates, particularly with regards to the age of flowering plants.