Microsoft AI Diagnostic System MAI-DxO Outperforms Doctors in Simulation Study

What happened

Microsoft has developed an AI diagnostic system called MAI-DxO (or DxO) that uses multiple AI agents to act like a medical panel [1]. According to a study published in the journal Science on Thursday, October 23, 2026, the system performed as follows in simulations using complex cases from the New England Journal of Medicine:

• MAI-DxO achieved 83% diagnostic accuracy [1, 2].
• Generalist physicians using reference materials achieved 36% accuracy [1, 2].
• Physicians working without reference materials achieved 22% accuracy [2].

In emergency department cases using data from Beth Israel Deaconess Medical Center, the system scored 80% accuracy, while human physicians averaged 47% [2]. Microsoft also created a benchmark called SDBench for these tests, which the company plans to open-source [2].

The system’s performance varies depending on the underlying model used. Microsoft tested DxO with models from OpenAI, Google, Anthropic, Meta, and DeepSeek [1]. The best results came from OpenAI’s GPT-5 and GPT-5.5 [1]. Using GPT-5 alone on complex cases resulted in 78.9% accuracy at an estimated cost of $8,400 per case; DxO increased accuracy by about four points and reduced the cost by approximately $3,000 [1]. With the GPT-5.5 model, released in April 2026, DxO increased accuracy from 72% to nearly 87% on complex cases [1]. On Gemini 2.5 Flash, DxO raised accuracy from 46.4% to nearly 69% [1].

Microsoft AI CEO Mustafa Suleyman stated the company is working with the Mayo Clinic to train a healthcare foundation model, with a release planned for next year [1]. Additionally, Microsoft plans to evaluate DxO in real-world hospital settings through a partnership with Beth Israel Deaconess Medical Center [1, 2]. Dominic King, vice president of Health at Microsoft AI, said these evaluations will likely begin in the next couple of months [2].

Who is affected

• Generalist physicians and internists: The study compared AI performance against these professionals [1].
• Beth Israel Deaconess Medical Center: This Boston hospital will participate in real-world clinical evaluations where consenting patients engage with DxO alongside standard treatment [1, 2].
• Clinicians: Suleyman noted that while AI may increase some workloads, it aims to reduce medical drudgery to allow more time for patient care [2].
• Healthcare providers: Microsoft aims to address the mismatch between medical expertise supply and demand [1].