Microsoft's MAI-DxO AI System Outperforms Doctors by Four Times in Diagnosis

Microsoft has officially launched MAI-DxO, a multi-agent medical diagnostic orchestration system, with findings published in Science. In complex case benchmarks from the New England Journal of Medicine (NEJM), the system achieved an 83% diagnostic accuracy rate, more than double that of general practitioners (36%) and described as four times higher in specific scenarios. This marks a critical leap for AI in clinical assistance, moving from reference to superiority.

The core innovation lies in its multi-agent collaboration mechanism, which simulates expert consultation processes. Rather than relying on a single model, MAI-DxO coordinates multiple AI agents to perform tasks such as inquiry, test recommendations, and diagnostic debates, approximating the comprehensive judgment of human multidisciplinary teams. Enhanced underlying foundation models (like GPT-5/5.5) drive performance, while the orchestration architecture amplifies these gains. Microsoft AI CEO Mustafa Suleyman emphasized that results based on the proprietary SDBench prove AI now possesses clinical capabilities approaching or exceeding human experts.

This breakthrough addresses global healthcare supply-demand imbalances. Positioned as a step toward "medical superintelligence," Microsoft predicts AI systems indistinguishable from humans in medical performance within years. This promises to lower diagnostic costs and free doctors to focus on patient care, reshaping service delivery.