RAXML-VI-HPC is a program designed for maximum likelihood (ML) phylogenetic inference, optimized for high-performance computing. It offers significant improvements in speed and efficiency compared to its predecessor, RAXML, through technical optimizations, a modified search algorithm, and the use of the GTR+CAT model as an alternative to the GTR+Γ model. The program has been parallelized with MPI for parallel bootstrapping and multiple inferences on distinct starting trees. Performance comparisons with other leading programs like GARLI, PHYML, IQPNNI, and MrBayes on real datasets with up to 6722 taxa show that RAXML requires less main memory and yields better trees in similar or less time. The GTR+CAT approximation is particularly effective in reducing memory consumption while maintaining or improving likelihood values. The program has been used to compute ML trees on large alignments, including one with 25,057 taxa and another with 21,822 taxa. Future work will focus on developing new methods for rapid bootstrapping to further enhance computational efficiency.RAXML-VI-HPC is a program designed for maximum likelihood (ML) phylogenetic inference, optimized for high-performance computing. It offers significant improvements in speed and efficiency compared to its predecessor, RAXML, through technical optimizations, a modified search algorithm, and the use of the GTR+CAT model as an alternative to the GTR+Γ model. The program has been parallelized with MPI for parallel bootstrapping and multiple inferences on distinct starting trees. Performance comparisons with other leading programs like GARLI, PHYML, IQPNNI, and MrBayes on real datasets with up to 6722 taxa show that RAXML requires less main memory and yields better trees in similar or less time. The GTR+CAT approximation is particularly effective in reducing memory consumption while maintaining or improving likelihood values. The program has been used to compute ML trees on large alignments, including one with 25,057 taxa and another with 21,822 taxa. Future work will focus on developing new methods for rapid bootstrapping to further enhance computational efficiency.