See this. I use OpenEvidence. It has access to full text from some of the major medical journals. But generalist models seem to outperform it. Not sure what is going on there.