Microsoft says its AI tool diagnoses conditions more accurately than doctors

Microsoft unveiled Diagnostic Orchestrator, a GPT-5-powered system that uses multiple AI agents to work through symptoms and order tests. It reached about 83% diagnostic accuracy on NEJM case reports versus 36% for generalist physicians, and 80% versus 47% on Beth Israel Deaconess emergency data. The findings were published in Science, and Microsoft plans to open-source its SDBench benchmark.

• About 83% accuracy on NEJM cases versus 36% for physicians with references
• 80% versus 47% on Beth Israel Deaconess emergency department cases
• The system runs multiple GPT-5 agents and diagnoses at lower estimated cost
• Hospital clinical testing is expected to begin in the coming months