INTRODUCTION: Differentiated service delivery (DSD) models for antiretroviral treatment (ART) have been scaled up in many settings in sub-Saharan Africa to improve client-centred care and increase service delivery efficiency. However, given the multitude of models of care currently available, identifying cost-effective combinations of DSD models that maximize benefits and minimize costs remains critical for guiding their expansion. METHODS: We developed an Excel-based mathematical model using retrospective retention and viral suppression data from a national cohort of ART clients (≥15 years) in Zambia between January 2018 and March 2022 stratified by age, sex, setting (urban/rural) and model of ART delivery. Outcomes (viral suppression and retention in care), provider costs and costs to clients were estimated from the cohort and published data. The base case reflects the outcomes observed in 2022 for all DSD models for each population sub-group. For different combinations of nine DSD models and over 1-year time horizon from the provider perspective, we evaluated the incremental cost-effectiveness ratio (ICER) per additional client virally suppressed compared to the 2022 base case. Deterministic sensitivity analyses were conducted on key input parameters. RESULTS: Among 125 scenarios evaluated, six were on the cost-effectiveness frontier: (1) 6-month dispensing (6MMD)-only; (2) 6MMD and adherence groups (AGs); (3) AGs-only; (4) fast track refills (FTRs) and AGs; (5) FTRs-only; and 6) AGs and home ART delivery. 6MMD-only was cost-saving compared to the base case, increasing retention by 1.2% (95% CI: 0.7-1.8), viral suppression by 1.6% (95% CI: 1.0-2.7) and reducing client costs by 12.0% (95% CI: 10.8-12.4). The next cost-effective scenarios, 6MMD + AGs and AGs-only, cost $245 per additional person virally suppressed, increased viral suppression by 2.8% (95% CI: 2.2-3.3) and 4.0% (95% CI: 3.5-4.0) and increased client costs by 20.1% (95% CI: 9.5-28.1) and 52.3% (95% CI: 29.868.7), respectively. ART cost and laboratory test costs were the most influential parameters on provider costs and the ICERs. CONCLUSIONS: Mathematical modelling using existing data can identify cost-effective DSD model mixes while ensuring all client sub-populations are considered. In Zambia, scaling up 6MMD to all eligible clients is likely cost-saving, with further health gains achievable by targeting sub-populations with selected DSD models.