Background Salvia splendens Ker-Gawler, scarlet or tropical sage, is a tender herbaceous perennial widely introduced and seen in public gardens all over the world. With few molecular resources, breeding is still restricted to traditional phenotypic selection, and the genetic mechanisms underlying phenotypic variation remain unknown. Hence, a high-quality reference genome will be very valuable for marker-assisted breeding, genome editing, and molecular genetics. Findings We generated 66 Gb and 37 Gb of raw DNA sequences, respectively, from whole-genome sequencing of a largely homozygous scarlet sage inbred line using Pacific Biosciences (PacBio) single-molecule real-time and Illumina HiSeq sequencing platforms. The PacBio de novo assembly yielded a final genome with a scaffold N50 size of 3.12 Mb and a total length of 808 Mb. The repetitive sequences identified accounted for 57.52% of the genome sequence, and 54,008 protein-coding genes were predicted collectively with ab initio and homology-based gene prediction from the masked genome. The divergence time between S. splendens and Salvia miltiorrhiza was estimated at 28.21 million years ago (Mya). Moreover, 3,797 species-specific genes and 1,187 expanded gene families were identified for the scarlet sage genome. Conclusions We provide the first genome sequence and gene annotation for the scarlet sage. The availability of these resources will be of great importance for further breeding strategies, genome editing, and comparative genomics among related species.