A multilingual benchmark of 1,886 vaccine-related questions found that large language models answered most items accurately ...