Inhibitory interneurons account for 20-30\% of neurons in the human neocortex (Markram et al. 2004) and separate into distinct cell-types (Gouwens et al. 2020, Lee et al. 2023). However, the functional role of these inhibitory cell-types is not yet well understood, especially not in the intact human neocortical circuit. Here we investigate the input-output transformations afforded by human inhibitory neurons as a function of cell-type compared to their rodent counterparts. To do so, we use multi-objective evolutionary optimization to develop approximately 4860 biophysically realistic and morphologically detailed single-neuron models of the four major inhibitory cell-types (PVALB, VIP, SST, and LAMP5 interneurons) from multimodal (single-cell transcriptomics, electrophysiology, and morphology) experimental data (Nandi et al. 2022). The biophysical models accurately reproduce electrophysiological features and differences across the cell-types, such as the frequency-current (F-I) relationship for current-clamp experiments. We simulate two different scenarios (direct injection to the soma vs. activation of dendritic synapses) and characterize their responses across cell-types and across human cortical layers. We find that PVALB neurons require the most input to fire under dc stimulation whereas they present the shortest decay time in their somatic responses during synaptic activation. We link these differences to specific morphoelectric properties and ion channel expression like the cortical depth profile of conductances (e.g. Ih, Im). Finally, we test how these cellular and cell-type differences unique to the human neocortex manifest themselves at the circuit level. To achieve this, we set up a biophysically faithful human neocortical circuit model, instantiate targeted perturbations in terms of ion channels and location, and test how circuit computations like coincidence detection and nonlinear amplification are impacted. Our simulations provide a methodology to systematically probe the computations different cell-types perform at both the single neuron and circuit level.