OpenAI's o3 model helps diagnose 18 rare-disease patients in NEJM AI study

A peer-reviewed study in NEJM AI reported that OpenAI's o3 model helped diagnose 18 children at Boston Children's Hospital who had rare neurodevelopmental and neuromuscular disorders. Researchers analyzed 376 genomes from previously undiagnosed patients, using the model to surface candidate causes that had eluded conventional workups, and described the outcome as a 'total game changer' for families who often endure years of diagnostic odysseys.

The result is a concrete, clinically validated win for frontier reasoning models in genomics — distinct from the hype-heavy benchmark claims that dominate the news cycle. It demonstrates value in a high-stakes domain where the cost of a missed diagnosis is enormous and where expert genetic interpretation is scarce. The study's structure (376 genomes, 18 diagnoses) gives a tangible hit rate that clinicians can scrutinize.

There's an irony in the timing: OpenAI announced this week that o3 will be retired from ChatGPT by August 26, 2026, as it consolidates its model lineup — even as research like this shows specific value in the very model being sunset. That highlights a tension between rapid model churn and the slower validation cycles of regulated, clinical use. Caveats from the genomics community will center on generalizability beyond a single hospital cohort and the need for expert oversight of AI-suggested diagnoses. Watch whether OpenAI maintains API access to o3 for research even after the consumer retirement.