The optimal timing of screening for diabetes mellitus (DM) among tuberculosis (TB) cases is unclear due to the possibility of stress hyperglycemia. We evaluated adult (18 years) pulmonary TB cases at treatment initiation as well as at 3 months, 6 months and 12 months. DM was identified by self-report (known DM) or glycated hemoglobin (HbA1c) 6.5% (new DM). Trends in HbA1c levels during treatment were assessed using non-parametric tests. Of the 392 participants enrolled, 75 (19%) had DM, 30 (40%) of whom had new DM. Of the 45 participants with known DM, respectively 37 (82%) and 40 (89%) received medication to lower glucose levels at treatment initiation and completion; one participant with new DM initiated glucose-lowering medication during follow-up. The median HbA1c level in participants with known, new and no DM was respectively 10.1% (interquartile range [IQR] 8.3-11.6), 8.5% (IQR 6.7-11.5) and 5.6% (IQR 5.3-5.9) at treatment initiation, and 8.7% (IQR 6.8-11.3), 7.1% (IQR 5.8-9.5) and 5.3% (IQR 5.1-5.6) at treatment completion (P < 0.001). Overall, 5 (12%) with known and 13 (43%) with new DM at treatment initiation had reverted to HbA1c < 6.5% by treatment completion (P = 0.003); the majority of reversions occurred during the first 3 months, with no significant reversions beyond 6 months. HbA1c levels declined with anti-tuberculosis treatment. Repeat HbA1c testing at treatment completion could reduce the risk of misdiagnosis of DM.