Microsoft AI Diagnostic System Achieves 83% Accuracy, Outperforms Doctors on Complex Cases

Microsoft AI Achieves Breakthrough Diagnostic Accuracy

Microsoft AI has demonstrated that its diagnostic system can outperform doctors on complex medical cases, achieving 83% accuracy compared to 36% for generalist physicians, according to a study published Thursday in Science journal1

1

2

2

. The AI diagnostic system, called MAI-DxO, represents what Mustafa Suleyman, Microsoft AI's CEO, describes as a major milestone toward medical superintelligence.

How the AI Tool for Diagnosing Diseases Works

MAI-DxO coordinates multiple AI agents that function like a panel of medical experts1

1

. These AI agents question patients, order tests, and debate diagnoses, mimicking the collaborative process physicians use to investigate challenging cases. The system was tested on scenarios drawn from The New England Journal of Medicine case reports and emergency department data compiled into Microsoft's own benchmark, SDBench, which the company plans to open-source2

2

.

Diagnostic Accuracy Across Different Scenarios

When diagnosing complex medical cases, primary care doctors and general internists working individually with reference materials were correct only 36% of the time, while those working without references managed just 22%2

2

. The AI diagnostic system's 83% accuracy represents more than double the performance of generalist physicians. On emergency department cases from Beth Israel Deaconess Medical Center in Boston, the system scored 80% accuracy compared to an average of 47% for human physicians2

2

.

Role of Advanced AI Models in Medical Diagnosis

Microsoft tested MAI-DxO with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, but achieved its best results using OpenAI's GPT-5 and GPT-5.51

1

. The study's data suggests much of the improvement comes from more powerful underlying AI models, with Microsoft's orchestration system adding gains on top. GPT-5 alone scored 78.9% on complex cases at approximately $8,400 per case, while MAI-DxO cut that cost by about $3,000 and raised accuracy by roughly four points1

1

.

Cost Optimization in AI-Powered Diagnostics

Microsoft measured MAI-DxO on both diagnostic accuracy and cost, tracking the estimated expense of running the AI on each case. With GPT-5.5, released in April 2026, the system increased accuracy from 72% to nearly 87% on complex cases at a slightly cheaper cost1

1

. The orchestration helps most when diagnosing complex medical cases and with less capable AI models. For instance, Gemini 2.5 Flash scored 46.4% independently but nearly 69% with MAI-DxO assistance1

1

.

Building Toward Medical Superintelligence

Dominic King, vice president of Health at Microsoft AI, emphasized that the biggest challenge in healthcare delivery is the growing mismatch between demand and supply of medical expertise1

1

. Mustafa Suleyman revealed that Microsoft is collaborating with Mayo Clinic to train a healthcare-specific foundation model, with release planned for next year1

1

. He believes this model will enable enormous leaps forward and predicts that healthcare models will get an order of magnitude better in capability within the next 24 to 36 months.

Real-World Clinical Validation Ahead

Microsoft acknowledges a long road remains before widespread deployment. Through a partnership with Beth Israel Deaconess Medical Center, the company will evaluate MAI-DxO's clinical decision support capabilities across different hospital settings1

1

. King indicated testing will likely begin in the next couple of months2

2

. While the system has only been tested on past medical cases, clinical validation will determine how well it performs in real clinical environments.

Future of AI in Healthcare Delivery

Suleyman, who co-founded DeepMind before joining Microsoft AI in 2024, stated he believes 2027 will bring a breakthrough for medical records, with highly accurate predictions about how medical records might unfold1

1

. Microsoft has outlined five additional areas where AI will need to master beyond diagnosis, including coordinating clinical care, making treatment decisions, and handling the human side of medicine2

2

. Suleyman insists human clinicians should and will always have a central role in patient care, even as AI capabilities advance. His hope is that AI will remove drudgery from medicine, freeing physicians to spend more time with patients.