Across the glacial-interglacial cycles of the late Pleistocene (~700 k.y.), temperature variability at low latitudes is often considered to have been negligible compared to changes in precipitation. However, a paucity of quantified temperature records makes this difficult to reliably assess. In this study, we used the Bayesian method CREST (Climate REconstruction SofTware) to produce a 790,000 yr quantified temperature reconstruction from a marine pollen record from southeast Africa. The results reveal a strong similarity between temperature variability in subtropical Africa and global ice volume and CO2 concentrations, indicating that temperature in the region was not controlled by local insolation, but followed global trends at these time scales, with an amplitude of ~4 °C between glacial minima and interglacial maxima. The data also enabled us to make an assessment of the impact of temperature change on pollen diversity, with results showing there is no link between glacial-age temperatures/CO2 and a loss of diversity in this record.