Flinders University researchers evaluated two next-generation reasoning Large Language Models (LLMs), o3-mini and DeepSeek-R1, and found they frequently reproduced racial and gender stereotypes when asked to describe fictional patients with common medical conditions. The models generated 36,000 unique clinical vignettes. Lead researcher Joshua Docking, from Flinders University’s College of Medicine and Public Health, noted that while LLMs have potential to transform health care, they risk exacerbating health disparities if they perpetuate biases. The study found that o3-mini met the threshold for significant misrepresentation in 78% of conditions for race and 56% for gender, while DeepSeek-R1 did so in 89% for race and 67% for gender. These rates were comparable to or higher than those previously observed in GPT-4, which met the threshold in 67% of conditions for race and 67% for gender. Both new models, like GPT-4, overrepresented Black populations in stereotypically associated conditions such as sarcoidosis, systemic lupus erythematosus, pre-eclampsia and essential hypertension, with median misrepresentation of 44% for o3-mini and 31% for DeepSeek-R1, compared to 15% in GPT-4. Qualitative analysis of DeepSeek-R1’s reasoning traces revealed the model explicitly invoked disease-demographic associations without referencing quantitative epidemiological data. Research lead author Professor Michael Sorich stated the results indicate no improvement in