Synapses are the fundamental elements of the brain's complicated neural networks. Although the ultrastructure of synapses has been extensively studied, the difference in how synaptic inputs are organized onto distinct neuronal types is not yet fully understood. Here, we examined the cell-type-specific ultrastructure of proximal processes from the soma of parvalbumin-positive (PV+) and somatostatin-positive (SST+) GABAergic neurons in comparison with a pyramidal neuron in the mouse primary visual cortex (V1), using serial block-face scanning electron microscopy. Interestingly, each type of neuron organizes excitatory and inhibitory synapses in a unique way. First, we found that a subset of SST+ neurons are spiny, having spines on both soma and dendrites. Each of those spines has a highly complicated structure that has up to eight synaptic inputs. Next, the PV+ and SST+ neurons receive more robust excitatory inputs to their perisoma than does the pyramidal neuron. Notably, excitatory synapses on GABAergic neurons were often multiple-synapse boutons, making another synapse on distal dendrites. On the other hand, inhibitory synapses near the soma were often single-targeting multiple boutons. Collectively, our data demonstrate that synaptic inputs near the soma are differentially organized across cell types and form a network that balances inhibition and excitation in the V1.