The paper describes how recent advances in large-scale deep reasoning language models, such as GPT and the open‑source DeepSeek, are expanding their potential use in ophthalmology. Currently, classical computer vision models are predominantly used to evaluate eye images, while text-based LLMs are mainly used for language tasks such as interpreting reports, preparing patient education materials, and summarizing electronic health records. Multimodal systems that combine visual inputs with reasoning have mainly been tested in simulated or retrospective conditions, such as in personalized planning. The direct clinical benefit of these solutions has not yet been demonstrated. Implementation in practice is made difficult by high computational demands, issues of privacy protection, biases, as well as problems with transparency and interpretability. Real clinical use is also complicated by system overload and unstable response times. The authors highlight the need for further research to focus on operational and ethical limits, adapting AI to ophthalmic workflows and ensuring that these tools remain helpful, fair and transparent decision-making partners. According to the article, prospective interventional studies are necessary before claims of improved patient outcomes can be made.