Pancreatic beta cells have been shown to be heterogeneous at multiple levels. However, spatially interrogating transcriptional heterogeneity in the intact tissue has been challenging. We developed an optimized protocol for single-molecule transcript imaging in the intact pancreas and used it to identify a sub-population of “extreme” beta cells with elevated mRNA levels of insulin and other secretory genes in mice. Extreme beta cells contain higher ribosomal and proinsulin content but lower levels of insulin protein in fasted states, suggesting they may be tuned for basal insulin secretion. They exhibit a distinctive intra-cellular polarization pattern, with elevated mRNA concentrations in an apical ERenriched compartment, distinct from the localization of nascent and mature proteins. The proportion of extreme cells increases in type 2 diabetic db/db mice, potentially facilitating the required increase in basal insulin. Our results thus highlight a sub-population of beta cells that may carry distinct functional roles along physiological and pathological timescales. We would like to apply our new technology of smFISH in the intact human pancreas to assess whether similar heterogeneity of extreme and non-extreme beta cells exists in human.