Microsoft AI Diagnostic System Scores 83% on Complex Medical Cases in Study

Microsoft AI’s multi-agent diagnostic system, MAI-DxO, accurately diagnosed complex simulated cases from the New England Journal of Medicine 83% of the time in a study published Thursday in Science, compared to 36% for primary care doctors and general internists working individually with reference materials, though specialists and physician teams who normally handle such cases were excluded from the comparison. The study's data indicates that most accuracy gains came from the underlying frontier models rather than the orchestration layer: OpenAI’s GPT-5 scored 78.9% on complex cases on its own at an estimated cost of about $8,400 per case, while adding DxO raised accuracy by about four points and reduced costs by about $3,000. With OpenAI's GPT-5.5, which released in April 2026, DxO lifted complex case accuracy from 72% to nearly 87% at a slightly lower cost, and boosted Google's Gemini 2.5 Flash from 46.4% to nearly 69%, though neither GPT-5 nor GPT-5.5 saw meaningful accuracy or cost improvements from DxO on emergency department cases. Microsoft AI CEO Mustafa Suleyman told TechCrunch the company is partnering with Mayo Clinic to train a healthcare foundation model planned for release in 2027, and will evaluate DxO in parallel with standard human care across emergency and urgent care settings at Boston's Beth Israel Deaconess Medical Center. "I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records." — Mustafa Suleyman, CEO of Microsoft AI.