Microsoft published research on an AI tool called Diagnostic Orchestrator. It diagnosed medical cases up to four times more accurately than human physicians. The system uses multiple AI agents to reason through symptoms and medical histories. It achieved 83% accuracy on NEJM cases versus 36% for generalist physicians. It scored 80% on emergency department cases versus 47% for human physicians. The tool also tended to reach diagnoses at a lower estimated cost. Microsoft plans to open-source its benchmark SDBench. The system is powered by OpenAI’s GPT-5. It outperformed single frontier AI models. Microsoft will begin testing the system in hospital settings with Beth Israel Deaconess Medical Center. The company aims to develop medical superintelligence. Commercialization is not yet timed. Microsoft emphasizes a responsible rollout due to regulatory requirements.
What changed
Human physicians diagnosed medical cases without AI assistance.
What it unlocks
Diagnosing medical conditions with higher accuracy and lower estimated cost than human physicians.
- 83% diagnostic accuracy vs 36% for generalist physicians
- 80% accuracy vs 47% for human physicians on emergency cases
- 4x more accurate than human doctors
What you need to act on it
- clinical validation
- real-world testing in hospital settings
Sources