RAxML-NG is a fast, scalable, and user-friendly tool for maximum likelihood phylogenetic inference. This supplement describes key improvements and new features compared to RAxML/ExaML. RAxML-NG supports 22 classical GTR-derived models, allowing different models for different partitions. It also supports flexible models for multi-state data, with up to 64 states and customizable encoding. Rate heterogeneity across sites is now handled with the FreeRate model, which does not rely on a priori rate distributions. RAxML-NG also supports three branch linkage models: linked, unlinked, and scaled (proportional), with the scaled model being the default for partitioned analyses. Per-rate Γ scalers are introduced to address numerical underflow issues in likelihood calculations. The tree search algorithm has been modified to ensure all promising subtree moves are considered. Transfer bootstrap (TBE) is implemented for improved branch support estimation. Phylogenetic terraces are automatically checked, and all trees on a terrace can be enumerated. RAxML-NG supports fine-grained parallelization with MPI and pthreads, and includes load balancing and checkpointing features. It also automatically detects hardware and estimates resource requirements. In evaluation, RAxML-NG outperformed other tools in search efficiency and inference speed, particularly on large datasets. It showed superior performance on taxon-rich datasets and provided faster inference times compared to RAxML, ExaML, and IQTree. RAxML-NG scales linearly with up to 1,024 cores and achieves superlinear speedups on DNA datasets. It is more efficient than ExaML, its predecessor, and provides better parallel efficiency. IQTree, while more stable, is slower and shows variability in results on taxon-rich datasets. Overall, RAxML-NG is a powerful and efficient tool for phylogenetic inference.RAxML-NG is a fast, scalable, and user-friendly tool for maximum likelihood phylogenetic inference. This supplement describes key improvements and new features compared to RAxML/ExaML. RAxML-NG supports 22 classical GTR-derived models, allowing different models for different partitions. It also supports flexible models for multi-state data, with up to 64 states and customizable encoding. Rate heterogeneity across sites is now handled with the FreeRate model, which does not rely on a priori rate distributions. RAxML-NG also supports three branch linkage models: linked, unlinked, and scaled (proportional), with the scaled model being the default for partitioned analyses. Per-rate Γ scalers are introduced to address numerical underflow issues in likelihood calculations. The tree search algorithm has been modified to ensure all promising subtree moves are considered. Transfer bootstrap (TBE) is implemented for improved branch support estimation. Phylogenetic terraces are automatically checked, and all trees on a terrace can be enumerated. RAxML-NG supports fine-grained parallelization with MPI and pthreads, and includes load balancing and checkpointing features. It also automatically detects hardware and estimates resource requirements. In evaluation, RAxML-NG outperformed other tools in search efficiency and inference speed, particularly on large datasets. It showed superior performance on taxon-rich datasets and provided faster inference times compared to RAxML, ExaML, and IQTree. RAxML-NG scales linearly with up to 1,024 cores and achieves superlinear speedups on DNA datasets. It is more efficient than ExaML, its predecessor, and provides better parallel efficiency. IQTree, while more stable, is slower and shows variability in results on taxon-rich datasets. Overall, RAxML-NG is a powerful and efficient tool for phylogenetic inference.