The prefrontal cortex (PFC) has been reported to play a key role in higher-order cognitive functions by dynamically adjusting its firing activity. Here, we studied the specific contributions of two cell subclasses, putative excitatory pyramidal (68%) and putative inhibitory interneurons (24%), within a sample of 147 single units recorded from the dorsolateral PFC of two macaque monkeys while they performed a Transitive Inference (TI) task. The TI task required learning the relationship between the adjacent items of an arbitrarily rank-ordered sequence in every session, i.e., A>B>C>D>E>F. Consequently, the animals were tested, including never-experienced non-adjacent item pairs (e.g., B-E). We previously demonstrated that the rank distance between items (Symbolic Distance; SDist) positively correlates with the performance accuracy during the test trials while modulating PFC activity. Here, we explored if the two cell subpopulations equally expressed this modulation while contributing to the progressive learning observed in the behavior during the test phase. We found that population activity from putative interneurons expressed a significant SDist modulation, with this encoding absent in the pyramidal subpopulation. Additionally, the performance increase from the first to the last half of the test trials was reflected in the neuronal activity in both cell types, with a significantly higher rate of change in the interneurons compared to pyramidal neurons. These results suggest a different role of the two classes of neurons during task execution and familiarization potentially related to changes in the ratio between inhibition and excitation in the local network.