BACKGROUND Although Traditional Chinese Medicine (TCM) is gaining global recognition, the availability of highly trained experts remains limited. The individualized and dynamically complex nature of TCM herbal formulations presents challenges unmatched in Western medicine. OBJECTIVE This study evaluates whether advanced, publicly accessible large language models (LLMs) can approximate the diagnostic reasoning and herbal prescription practices of experienced TCM doctors, using a standardized migraine case from a published report. METHODS Nine LLMs (eight reasoning-enabled “thinking” models and one non-thinking baseline) were prompted through publicly available interfaces using a structured five-task format (Western diagnosis, TCM diagnosis, treatment principle, herbal prescription, and preventive care). No fine-tuning or external knowledge bases were applied. For comparison, diagnoses and prescriptions from three TCM doctors, including the original case author, were generated using the same prompt. Thirty-two expert raters, blinded to response source, independently scored all outputs across five evaluation dimensions. RESULTS Most reasoning-enabled LLMs achieved performance comparable to that of senior TCM physicians, with several models (e.g., GPT-o3, Qwen-3, Gemini-2.5 Pro) receiving significantly higher scores than one or more physicians in specific tasks. By contrast, the non-thinking baseline model (Aya Expense 32B) performed substantially worse. CONCLUSIONS Advanced LLMs may have reached a “Turing point” in TCM, rivaling expert performance. While promising for addressing practitioner shortages, these advances highlight the urgent need for careful regulation and oversight.
Yan et al. (Tue,) studied this question.