Synaptic plasticity underlies our ability to learn and adapt to the constantly changing environment. The phenomenon of synapses changing their efficacy in an activity-dependent manner is often studied in small groups of neurons in vitro or indirectly through its effects on behavior in vivo. Investigating synaptic plasticity at an intermediate microcircuit level relies on simulation-based approaches, which offer a framework to reconcile fragmented and sparse experimental observations. Since Hebb's initial postulate, theoreticians have provided valuable insights about the role of cell assemblies, strongly interconnected groups of co-firing neurons, in learning and memory. However, most of these studies were limited in their scale, biological realism, and therefore generality. To overcome these limitations, we further improved and validated our previously published large-scale cortical network model featuring short-term plasticity and equipped it with a recently developed calcium-based model of long-term plasticity between excitatory cells. We calibrated the network to mimic an in vivo state characterized by low synaptic release probability and low-rate asynchronous firing and exposed it to ten different stimuli. By virtue of the model's non-random, biorealistic connectivity we could detect cell assemblies from the stimulus-evoked spiking activity of 186,665 neurons even in the naïve circuit, before the long-term plastic changes. This detection employed a combination of methods established by experimentalists. Leveraging the in silico nature of our setup, we then studied how the structure of synaptic connectivity underlies assembly composition ranging from feedforward thalamic innervation to intricate high-dimensional network motifs of the recurrent connectivity. Notably, we found that long-term plasticity sparsely and specifically strengthened synapses between cell assemblies: among 312 million synapses, only 5% experienced noticeable plasticity in 10 minutes of biological time and cross-assembly synapses underwent three times more changes than average. As our model neurons featured realistic morphologies and dendritic ion channels, we could also investigate how nonlinear dendritic processes influence assembly membership and the effects of long-term plasticity on synapses forming spatial clusters on postsynaptic dendrites. A comparative analysis of the network's responses to the different stimuli before and after the long-term changes revealed a network-level redistribution of efficacy from the superficial to the deep cortical layers. This shift led to prolonged stimulus-specific responses and more assemblies activating exclusively for a single pattern. In summary, using a state-of-the-art, bottom-up model of the cortical microcircuit we found sparse and specific plastic changes that reconfigured network dynamics while preserving its stability.