De novo peptide sequencing can annotate mass spectra with their originating peptides without the requirement for a protein sequence database or needing to know the composition of a sample a priori. We have developed Casanovo, a state-of-the-art de novo peptide sequencing solution based on the popular transformer deep learning architecture. Casanovo significantly outperforms alternative de novo peptide sequencing tools. We demonstrate its state-of-the-art performance on a nine-species benchmark dataset and highlight advances in analyzing immunopeptidomics and metaproteomics data.