Breaking

Microsoft's AI Diagnostic System Achieves 85% Accuracy on Complex Medical Cases in Science Study

October 22, 2026 at 14:49 EDT

Foundation Models
Healthcare
Research & Papers

**Microsoft researchers published findings in *Science* showing their Sequential Diagnosis language model system achieves 85.5% accuracy on rare and complex medical cases from the New England Journal of Medicine, significantly outperforming experienced physicians who scored 20% on the same cases.** The study represents a major benchmark for AI diagnostic reasoning, demonstrating capabilities on cases designed to challenge expert clinicians.

OCTOBER 22 · MICROSOFT RESEARCH

AI Diagnostic System Outperforms Doctors 4× on Hard Cases

Microsoft's language model orchestration system achieved 85.5% diagnostic accuracy on rare and complex medical cases, far exceeding experienced physicians' performance.

85.5%

AI System Accuracy

20%

Physician Accuracy

Performance Multiplier

1×

Physician

4.25×

AI System

On 304 complex medical cases from NEJM, Microsoft's system dramatically outperformed experienced physicians.

✓ With References

80%

ED cases, AI system accuracy

⚠ Physician Baseline

47%

ED physician accuracy

Key Finding

Most of the improvement comes from underlying foundation models, not from Microsoft's orchestration system itself.

The system coordinates multiple AI agents, but the dramatic performance gains primarily reflect advances in GPT-5 and o3 foundation models across the LLM landscape.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.