Diabetic Retinopathy (DR) stands as a severe diabetes complication levels, among the major contributors to worldwide visual impairment and preventable blindness. The microvasculature damage of the retina that occurs from sustained high blood glucose levels causes this condition. The outcome leads to permanent blindness when treatment is not provided. The blending of insufficient preprocessing and intricate image changes in retinal images causes current detection algorithms to experience decreased accuracy together with substandard feature representation and reduced sensitivity. The research establishes Artificial Bee Colony Optimization based Hybrid Vision Multi Perceptron (ABC-ViM) as a new model for DR detection that uses a Multi-Layer Perceptron (MLP) for classification and a Vision Transformer (ViT) model for enhancing feature learning. The ABC algorithm functions to optimize classification boundaries through which the model achieves better decision-making performance and increased reliability. The hybrid ABC-ViM model generates 96.4% accuracy through its performance while achieving 95% precision and 96% recall along with a 95.5% F1 score.