Statistical and biochemical studies have revealed nonrandom patterns in codon assignments. The canonical genetic code is known to be highly efficient in minimizing the effects of mistranslational errors and point mutations, since it is known that, when an amino acid is converted to another due to error, the biochemical properties of the resulted amino acid are usually very similar to those of the original one. In this study, we have taken into consideration both relative frequencies of amino acids and relative gene copy frequencies of tRNAs in genomic sequences in order to introduce a fitness function which models the mistranslational probabilities more accurately in modern organisms. The relative gene copy frequencies of tRNAs are used as estimates of the tRNA content. We also altered the rule previously used for the calculation of the probabilities of single base mutation occurrences. Our model signifies higher optimality of the genetic code towards load minimization and suggests the presence of a coevolution of tRNA frequency and the genetic code.