OpenAI's o1 can compress the years-long process of diagnosing rare genetic conditions, according to a demonstration from geneticist Catherine Brownstein.
Brownstein walked through how o1 approaches rare medical cases - the kind where patients spend years cycling through specialists before anyone names their condition. The model reasons through clinical and genetic clues to surface candidate diagnoses at a pace traditional workflows can't match. OpenAI published the demonstration directly, positioning o1 as a tool for high-stakes medical reasoning.
Rare disease diagnosis is one of medicine's worst bottlenecks, with the average patient waiting years before a correct answer - a grind clinicians call the diagnostic odyssey. A reasoning model that can quickly narrow the field of possible diagnoses has genuine value, even as a first-pass filter for specialists to interrogate rather than a replacement for their judgment.
Still, this is OpenAI's own showcase, not a peer-reviewed study - and the distance between a compelling demo and a clinically validated tool is exactly the gap that buried IBM Watson for Oncology.