The Saturn study showed that popular AI models regularly make mistakes when answering financial questions. In complex scenarios requiring multiple calculations, the share of inaccurate answers reached 88%, with some models erring on almost every such request.