Mutations are the primary source of all genetic variation. Knowledge about their rates is critical for any evolutionary genetic analyses, but for a long time, that knowledge has remained elusive and indirectly inferred. In recent years, parent-offspring comparisons have yielded the first direct mutation rate estimates. Here, we estimated the mutation rate for the guppy (Poecilia reticulata), a model species in ecoevolutionary studies. We sequenced 4 parents and 20 offspring and screened their genomes for de novo mutations. The initial large number of candidate de novo mutations was analysed with a supervised machine learning approach to remove false-positive results. Additionally, candidate de novo mutations were validated using Sanger sequencing, and positively verified variants were used to estimate the mutation rate. The guppy mutation rate (µ = 3.44 × 10-9 per site per generation) is among the lowest directly estimated mutation rates in vertebrates. Similarly, low estimates were obtained for two other teleost fishes. We discuss potential explanations for such a pattern, as well as the utility of machine-learning approaches for standardized across experiments approach to estimate mutation rates.