OpenEvidence launches 4 medical AI models

As AI becomes embedded across the care continuum, developers are competing to create the fastest and most accurate models to support the complexity of clinical decision-making.

AI-powered medical knowledge platform OpenEvidence's latest entry in this race is a group of medical AI models. These include one AI model called Darwin, which is built to support clinical reasoning in the most challenging clinical cases and research.

The model is currently available as a research preview and can only be accessed by institutional partners, research collaborators and academic researchers for specific applications. OpenEvidence evaluated the model against available foundation models on four independent medical AI benchmarks, including MedQA.

The MedQA dataset contains U.S. Medical Licensing Examination-style questions drawn from professional board examinations. The model answered all 660 questions correctly. In comparison, Claude Fable 5 scored 99.7%, GPT-5.6 Sol scored 99.1% and Gemini 3.7 Flash scored 99.2%.

Additionally, the model outperformed the other three on the MedXpertQA, HealthBench Professional and NOHARM benchmarks.

However, OpenEvidence included a caveat for these results. The benchmarks test the models working alone, without a human in the loop.

"That does not match the reality of how clinical decision support tools are used today," a blog post on the company's website reads. "In practice these tools are an aid to clinical judgment rather than a substitute for it, and an answer is 'good' if it helps a physician make a better decision. Feng et al. (2026) set out this standard, and we believe it is the one clinical tools should ultimately be judged against."

As AI's abilities accelerate, the question of how much the tool should be allowed to do remains. Clinical AI experts previously told Healthtech Analytics that even if AI shows great potential in performing clinical reasoning tasks, the goal should be optimal human-AI collaboration.

In addition to Darwin, OpenEvidence has released three other models on its platform: Osler, Sackett and Snow. Osler is an upgrade to the model that currently powers the clinical intelligence platform and will become the new default model. It is the fastest of the three, providing answers to clinical questions at the point of care in under five seconds, the press release stated.

Sackett provides more extensive answers than Osler because it completes more rounds of searching and is likely to ask for clarification. Responses will take around 30 seconds to generate.  

Snow conducts multiple rounds of searching and provides a comprehensive report. With a five-minute turnaround time, this model is intended for complex cases involving differentials, competing comorbidities or questions where the evidence is not established, according to the company.

The moves come as OpenEvidence advances its offerings at an accelerated rate. In the first six months of the year, the company has launched a quality grading feature for its AI-powered answers, added an AI tool that can predict whether a patient has structural heart disease and unveiled a coding suggestion capability.

Anuja Vaidya has covered the healthcare industry since 2012. She currently covers healthcare IT and innovation, including artificial intelligence, digital healthcare, EHRs and interoperability.