A biologically grounded computational model built to mimic real neural circuits, not trained on animal data, learned a visual categorization task just as actual lab animals do, matching their accuracy, variability, and underlying neural rhythms.