Mathematical models are widely used to evaluate HIV programme performance, predict future HIV trends and understand variations in HIV risk. However, most HIV models used in the African context are developed by international modelling teams in high-income countries. The Thembisa model is a rare example of a locally-developed model, which is extensively used to support local programme evaluation, budgeting, planning and policy decisions. Since version 1.0 of the model was published in 2014, the model has been regularly updated to incorporate new data and research, and in consultation with local partners, to address key policy questions. This paper provides an overview of the model structure and reviews major changes in the evolution of the model. Key model changes include the extension of the model to provincial level in 2016, the modelling of transmission between men who have sex with men in 2017, tuberculosis in 2022-23, long-acting injectable pre-exposure prophylaxis in 2025 and HIV transmission through needle sharing in 2026 (version 5.0). The community of users has also expanded over time, most significantly with the first South African HIV Investment Case in 2015-16, UNAIDS and the National Department of Health's adoption of Thembisa as the official source of South African HIV model estimates in 2017, and the integration of Thembisa into the Naomi district-level HIV estimation process from 2020. Major future challenges are to maintain relevance in the face of a changing funding landscape, with decisions around optimization of HIV programmes increasingly dependent on broader disease priorities and health system constraints.