Comparative cost of care recommended by large language models versus treating physicians.

Comparing AI-generated care recommendations to physician recommendations can reveal differences in proposed resource use, but lower estimated costs do not necessarily indicate better care. This study compared the costs of care recommended by 24 LLMs with those recommended by treating physicians and found the AI-recommendations cost three times as much per visit. Further evaluation should examine patient outcomes, omitted care, and the generalizability of the cost estimates associated with AI testing and referral suggestions.