This research presents a comprehensive evaluation of reasoning-focused Large Language Models in ophthalmology, testing four cutting-edge models across 5,888 medical questions and scenarios.

• DeepSeek-R1 demonstrated superior performance (86.8% accuracy), outperforming human ophthalmology residents
• All models showed strength in foundational knowledge but struggled with complex clinical reasoning
• Significant performance gaps exist between models, with OpenAI o1 trailing at 74.7% accuracy
• Models excelled in ophthalmology basics but showed weaknesses in clinical decision-making

These findings are crucial for medical implementations as they highlight both the promise and limitations of AI in specialized healthcare settings, suggesting careful validation is needed before clinical deployment.

Benchmarking Next-Generation Reasoning-Focused Large Language Models in Ophthalmology: A Head-to-Head Evaluation on 5,888 Items