A review published in *Eye & ENT Research* describes the development of ophthalmic foundation models that use multimodal reasoning to analyze retinal photographs, optical coherence tomography, scanning laser ophthalmoscopy, clinical records, and other patient data in a single framework, while warning that impressive laboratory results have not yet translated into sufficient real-world evidence. Traditional convolutional neural networks (CNNs) have achieved strong performance in narrowly defined tasks like identifying diabetic retinopathy, glaucoma, and age-related macular degeneration, but they require large collections of expert-labeled images and are usually designed for one specific task or modality, limiting generalization. Foundation models use self-supervised learning on very large datasets, enabling few-shot and zero-shot prediction. The review traces a rapid progression: RETFound produced generalizable representations for retinal imaging but was not designed to fuse multiple data types; VisionFM supported eight ophthalmic imaging modalities; EyeCLIP brought together 11 imaging modalities and clinical text; EyeFM combined five imaging modalities with a large language model; and MIRAGE concentrated on paired optical coherence tomography and scanning laser ophthalmoscopy data. Across reported evaluations, multimodal systems generally outperformed comparable single-modality models, particularly when only a small number of labeled examples were available, because different forms of evidence can capture different aspects of disease.