Medical AI reaches $15 billion valuation, but clinical effectiveness remains in question

A startup combining AI and medical literature

OpenEvidence uses artificial intelligence to research and synthesize medical knowledge, providing answers supported by expert references. Following a $250 million funding round in September, the company was valued at $15 billion. The platform claims that 65% of US doctors use it, a figure stated by the company itself and not verified by third parties.

Testing at a tech conference

The tool was tested at Nvidia’s GTC conference, held from March 16 to 19, 2026, in San Jose, USA. The interaction follows a model similar to ChatGPT: users type questions into a chat interface and receive answers in natural language, supplemented by medical references and anatomical images illustrating the concepts discussed.

Integration with electronic health records

In March, Mount Sinai announced a partnership to integrate OpenEvidence directly into its electronic health record system, making the tool available to doctors, nurses, and pharmacists. The institutional version is free, and the company guarantees that clinical data from this version is not used to train its models. The goal is to enable professionals to consult evidence without leaving the clinical environment.

Content and sources

The platform’s knowledge base includes agreements with publications such as the New England Journal of Medicine and JAMA, as well as guidelines from the National Comprehensive Cancer Network. Despite access to these sources, users must still critically evaluate whether each answer applies to the specific case they are handling.

Scientific reviews highlight limitations

A study published in June in Nature Medicine compared OpenEvidence and UpToDate Expert AI with three general-purpose models. It analyzed 500 medical knowledge questions, 500 items of alignment with clinical criteria, and 100 real questions formulated by doctors, reviewed by 12 specialists. The general models outperformed the others in all tests. A correspondence in the same journal, published in September, questioned the scope of the conclusions due to the study’s methodological limitations.

Another review of 11 studies, published in August in npj Digital Medicine, described promising results but identified small sample sizes and heterogeneous methodologies. The authors called for prospective studies in real clinical settings and the development of standardized evaluations to track the platform’s ongoing evolution.

Current situation in Brazil

In Brazil, initiatives such as the Capricórnio Project are already working to develop similar tools for evidence research in oncology in Paraná. However, OpenEvidence’s institutional offering is limited to organizations serving verified professionals in the United States, according to information on its commercial website.