ArticleCureus2025
Performance of Large Language Models and Top-Decile Doctors on an Undergraduate Ophthalmology Examination.
Article in Cureus, 2025. The graph could read no effect estimate from its abstract, so it casts no vote on the map. Not yet cited in PubMed.
What it found
Each row is one number read from the abstract, on the scale the paper reported it, with its interval. Left of the dashed line favours the treatment, right favours the comparator. Under each row is the sentence it came from. New to these charts? A ten-minute tutorial.
The abstract states no effect estimate the extractor could read, or names no intervention and outcome on the map, so this paper lights no cell and moves no belief. It is still indexed, cited and linked below.
The trial behind it
Trials whose registry record cites this paper, or whose number appears in the abstract. A trial that started after this paper was published is citing it as background, not reporting it.
Neither the registry nor the abstract names a trial number. If this is a trial report, that itself is worth knowing.
Who cites it
0 citing papers in PubMed.
No citing paper in PubMed yet.
Corrections and comments
PubMed lists nothing against this paper. Absence here is not a guarantee, only a check that was made.
Authors and funding
2 authors.
Funding
No grant is acknowledged in the PubMed record.
Abstract
introductionThe rapid advancements of artificial intelligence (AI) and large language models (LLMs) have led to their increasing use in medical education and clinical practice. While studies demonstrate their proficiency in passing medical examinations, it remains unclear how their performance compares to that of human experts on a challenging, domain-specific examination.
methodsThis study compared the performance of four contemporary LLMs (ChatGPT-5 (OpenAI, San Francisco, CA), Claude Sonnet 4 (Anthropic, San Francisco, CA), Gemini Pro (Google, Mountain View, CA), and Perplexity (Perplexity AI, San Francisco, CA)) with a cohort of three doctors who had all recently scored in the top 10% of candidates on the Duke Elder Undergraduate Prize Examination. Participants were assessed on a 100 multiple-choice question (MCQ) examination that mimicked the style and difficulty of the prize exam. A two-tailed independent samples t-test was used for a primary comparison of the pooled LLM and pooled doctor cohorts, while a one-way ANOVA with Tukey's HSD post-hoc analysis was used for a secondary comparison of the four individual LLMs and the pooled doctor cohort.
resultsThe pooled LLM cohort, with a mean score of 85.50%, performed similarly to the pooled doctor cohort, which scored 75.67%. A primary analysis revealed no statistically significant difference between these groups (p = 0.0533). However, a secondary analysis found a significant difference among the individual LLMs and pooled doctor cohorts (p = 0.0370), with a post-hoc analysis showing that Claude Sonnet 4 (91.00%) significantly outperformed the pooled doctors (p = 0.0268). A sub-analysis on the optics subsection of the exam showed no significant difference between the pooled LLM cohort and the pooled doctor cohort (p = 0.5072, and no significant difference between individual LLMs and the pooled doctor cohort (p = 0.3545, respectively).
conclusionThe findings suggest that popular LLMs, particularly Claude Sonnet 4, have achieved a level of performance on par with high-performing doctors for a challenging ophthalmology examination. This study highlights the immense potential of LLMs as valuable educational tools in ophthalmology, while also underscoring the importance of their continued evaluation against rigorous clinical benchmarks and human oversight in practice.
Indexed as
Identifiers
What OpenQuestion holds
Registered trials
Read under generation 80e0d062 · epoch 390. Bibliography from PubMed, PubMed Central and OpenAlex; grants from NIH RePORTER; trial links from ClinicalTrials.gov; estimates, votes and beliefs from the OpenQuestion graph.