Acute myeloid leukaemia (AML) is an aggressive malignancy characterized by the accumulation of transformed immature myeloid blasts. Although standard chemotherapy routinely induces disease remission, most AML patients ultimately relapse with resistant disease. A notable exception is the AML subtype known as acute promyelocytic leukemia, where retinoic acid induces leukaemia maturation and transient remission, and is frequently curative in combination with arsenic trioxide. Recently agents including mutant IDH1/2 inhibitors and DHODH inhibitors have been shown to induce maturation and regression of other AML subtypes, sparking renewed interest in AML differentiation therapy. To examine differentiation therapy dynamics in vivo we have generated a novel mouse AML model driven by reversible knockdown of PU.1, a myeloid transcription factor functionally compromised in >50% of AML cases. RNAi-mediated PU.1 knockdown results in disseminated AML in vivo, and subsequent restoration of endogenous PU.1 in established AML triggers synchronous differentiation of leukemic blasts followed by widespread disease clearance. Despite near-complete remission upon PU.1 restoration, mice reproducibly relapse with immature AML within months. Notably, in vivo AML treatment reveals that one week after differentiation stimulus leukemic blasts mature into two myeloid lineages with distinct immunophenotype and morphology. AML-derived SSClowLy6G+ cells resembling neutrophils initially predominate but are eradicated in vivo within two weeks, consistent with the rapid turnover of normal neutrophils. In contrast, high resolution flow cytometry and imaging indicate that mature AML-derived SSChighF4/80+SigF+ eosinophil-like cells persist at low numbers in specific organs during disease remission and seed relapse within the spleen. In mice transplanted with AML blasts lacking the essential eosinophil lineage transcription factor GATA1, in vivo PU.1 restoration triggers neutrophil but not eosinophil lineage differentiation, thereby effectively eliminating residual disease. These results demonstrate that AML differentiation therapy can produce long-lived sublineages of mature AML-derived cells from which relapse can originate. Understanding the multilineage potential of AML blasts in individual patients may inform strategies that preclude or eradicate mature AML-derived cells to improve differentiation therapy outcomes.