Background: Diabetic retinopathy (DR) is one of the most prevalent microvascular complications of diabetes mellitus and a leading cause of visual impairment worldwide. However, clinically available, minimally invasive, and convenient tools for early DR screening and disease stage stratification remain inadequate. This study aimed to identify serum metabolic biomarkers for DR stage discrimination and to characterize metabolic alterations associated with DR onset and progression. Methods: Serum samples from four groups—healthy controls (HC), diabetic patients without retinopathy (NDR), patients with mild nonproliferative DR (NPDR), and patients with moderate-to-advanced DR (MADR)—were analyzed using untargeted metabolomics. Three pairwise comparisons were designed to identify key transition points in DR progression. Candidate biomarkers were screened by identifying differential metabolites and further optimized using logistic regression analysis. Results: Stage-specific metabolic reprogramming occurred during DR progression, and lipid metabolism dysregulation persisted throughout the disease course. The candidate biomarker panels achieved cross-validated area under the curve (AUC) values of 0.918, 0.800, and 0.893 for distinguishing NPDR from HC, NPDR from NDR, and MADR from NDR, respectively, demonstrating stable and robust diagnostic performance. Moreover, the identified biomarkers were significantly correlated with key clinical parameters of DR. Conclusions: This study provides promising blood-based candidate biomarkers and novel metabolic insights for early screening and progression monitoring of DR.