Abstract Background Gestational diabetes mellitus (GDM) is a common obstetric complication worldwide that seriously threatens maternal and fetal health. As the number of women conceiving through in vitro fertilization (IVF) continues to rise, this population is recognized as being at an elevated risk for GDM. However, there is still no consensus on the early prediction of GDM in IVF patients due to the lack of reliable biomarkers. Methods We compared the first-trimester serum cytokine and antibody profiles in 38 GDM women and 38 matched controls undergoing IVF treatment, based on the extensive human biobank of our large‑scale assisted reproductive cohort platform. The 76 samples were divided into a training set (n = 53) and a testing set (n = 23) using a 7:3 ratio, and five diverse machine-learning models for predicting GDM were constructed. Results By combining the top five differentially expressed first‑trimester serum biomarkers [including total immunoglobulin (Ig)G, total IgM, interleukin (IL)-7, anti‑phosphatidylserine (aPS)-IgG immune complexes (IC), and IL-15], a novel early prediction model was constructed, which achieved superior predictive value [area under the curve (AUC) and 95% confidence interval (CI) 0.906 (0.840-0.971), with a sensitivity of 75% and a specificity of 94.7%] for GDM development. The eXtreme Gradient Boosting (XGBoost) model achieved an AUC of 0.995 (95% CI: 0.995-1.000, P