Benchmarking LLM chatbots’ oncological knowledge with the Turkish Society of Medical Oncology’s annual board examination questions

Abstract Background Large language models (LLMs) have shown promise in various medical applications, including clinical decision-making and education. In oncology, the increasing complexity of patient care and the vast volume of medical literature require efficient tools to assist practitioners. However, the use of LLMs in oncology education and knowledge assessment remains underexplored. This study aims to evaluate and compare the oncological knowledge of four LLMs using standardized board examination questions. Methods We assessed the performance of four LLMs—Claude 3.5 Sonnet (Anthropic), ChatGPT 4o (OpenAI), Llama-3 (Meta), and Gemini 1.5 (Google)—using the Turkish Society of Medical Oncology’s annual board examination questions from 2016 to 2024. A total of 790 valid multiple-choice questions covering various oncology topics were included. Each model was tested on its ability to answer these questions in Turkish. Performance was analyzed based on the number of correct answers, with statistical comparisons made using chi-square tests and one-way ANOVA. Results Claude 3.5 Sonnet outperformed the other models, passing all eight exams with an average score of 77.6%. ChatGPT 4o passed seven out of eight exams, with an average score of 67.8%. Llama-3 and Gemini 1.5 showed lower performance, passing four and three exams respectively, with average scores below 50%. Significant differences were observed among the models’ performances (F = 17.39, p