Current practices for model selection (e.g., AIC, BIC) are aimed at finding the underlying model, but don't always produce the most accurate phylogeny. ModelTeller is a machine-learning based application that predicts the best substitution model for phylogenetic reconstruction, with an emphasis on branch-lengths accuracy!
Output:
    1. The best predicted model and the ranking of alternative models, in case that you cannot use the best model in downstream analysis.
    2. The maximum-likelihood tree, reconstructed using the best model.
    3. The feature contribution analysis to learn the weight of every feature in the prediction.

Select the best model for branch-lengths estimation
Use a fixed GTR+I+G topology -- Recommended!
Use a user defined topology