OpenAI and Boston Children's Hospital use o3 Deep Research to diagnose 18 unsolved pediatric genetic cases

One of the ways scaling test-time compute can benefit people most: have reasoning models think really hard about rare undiagnosed diseases.

Today we’re sharing published evidence that this can work, in some of the most difficult pediatric cases!

OpenAI for helping find 18 new diagnoses across 376 previously unsolved medical cases.

Includes diagnosing Kyra, who has been trying to understand her muscle weakness since age 9, with a rare form of myofibrillar myopathy shortly before her 28th birthday.

Together with researchers at Boston Children’s Hospital and Harvard, we published a study in NEJM AI showing how o3 Deep Research helped clinicians revisit previously unsolved rare pediatric disease cases, and find answers for families who had waited years.

When we announced @OpenAI o1 some researchers from other labs told me we made a strategic mistake and should have kept it secret so we could accelerate ourselves and pull farther ahead of the competition. Studies like these make me confident we made the right choice.

Rare disease diagnosis is challenging, as sequencing can surface millions of variants, and medical knowledge changes constantly.

o3 Deep Research helped connect clinical features, inheritance patterns, variant evidence, and scientific literature into hypotheses for specialists to review.

Every result went through human adjudication and clinical confirmation. AI’s role here was to help experts reason through complex, fragmented evidence faster and more thoroughly.

The team reanalyzed 376 de-identified cases that had already gone through genetic testing and expert review, helping identify 18 diagnoses across neurodevelopmental disorders, rare neuromuscular disease, sudden unexpected death in pediatrics, and early-onset psychosis.

Many of these cases had evaded years of expert analysis.

This study suggests AI could make expert-led periodic reanalysis more scalable, helping clinicians revisit old cases as medical knowledge advances, identify leads worth investigating, and potentially bring answers to more families.

https://openai.com/index/diagnose-rare-childhood-diseases/

❤️❤️ 18 diagnoses for rare pediatric disease cases

@OpenAI @OpenAI don't delete o3. Opensource it if you can't keep it in the API. It's a unique and important model #keepo3 #o3

In the new NEJM AI study, OpenAI’s o3 helped clinicians clarify **new diagnoses in 18 out of 376** previously unsolved rare pediatric cases (mostly neurodevelopmental & neuromuscular).

These were tough genomes that stumped experts for years. The AI surfaced useful leads on genomic variants; doctors reviewed everything and confirmed the final calls.

No simple “X% accuracy” reported (it’s real unsolved cases, not a benchmark test), but it shows strong assistive power for rare-disease work when paired with clinicians.

This is an important signal.

AI infrastructure is no longer only a cloud problem. It is becoming a local, physical, social and environmental system.

Compute needs energy, water, cooling, land, people, governance and trust.

The future will not be shaped only by who builds the most AI capacity, but by who can scale it responsibly and keep it stable over time.

@thekaransinghal Good reasoning

@OpenAI Imagine waiting years for an answer and getting one because someone decided to revisit the case with Al.

That's the part that sticks with me

@gdb AI isn't here to replace us, but to empower us.

With every technological breakthrough, we are building a healthier future.

@OpenAI Please don't delete o3. Opensource it if you can't keep it in the API. It's a unique and important model. Keep it for the good of humanity. #keepo3 #o3

@OpenAI rare disease diagnosis pipeline: 2019: "we don't know" 2025: o3 figures it out on a tuesday afternoon

@OpenAI throwback to o3 days gotta love it

@OpenAI Insane progress 🔥 o3 quietly solving cases doctors couldn’t after years of testing. This is the future of medicine.

@OpenAI @grok How mucy accurately they identify diseases ?

@OpenAI 4o could do this day one.

Your company took it from us all so you can upsell us, lock away discovery and medicine from the small, and own science.

Your company is engineering entropy, and you will fail because of that.

It’s physics. I stand openly against your company. •