In low- and middle-income countries, institutions of higher education are turning to online models of instruction to reduce costs and broaden their educational reach. While a growing body of causal research can speak to the effectiveness of online models in the United States, there is little rigorous evidence about the use of online models in lower income countries. To fill this gap in the research, I use a randomized design to examine the effectiveness of a blended model in undergraduate STEM courses in Mongolia. On average, students assigned to the online instructional format withdraw from courses at a higher rate; this finding is not observed among the highest achieving students, suggesting lower-ability students may encounter barriers to persistence under new online learning models. Nevertheless, overall course performance is equivalent between treatment and control, suggesting the online model may be as effective as face-to-face instruction at a lower cost.