Microsoft AI on October 22, 2026 published peer-reviewed results in Science for MAI-DxO, its Diagnostic Orchestrator, reporting that the system reached the correct diagnosis far more often than generalist physicians on a benchmark built from difficult medical cases. The company is careful to frame the work as a benchmark result, and the study's own authors stress that it was not conducted in real clinical care.
The test bed is the Sequential Diagnosis Benchmark, or SDBench, which turns 304 complex case records from The New England Journal of Medicine and a set of more typical emergency-department admissions from the MIMIC-IV-ED dataset into step-by-step encounters. A physician or AI system starts with a short case summary and must decide what questions to ask and which tests to order, with each request carrying an estimated cost. The benchmark scores both whether the final diagnosis is right and how much the workup would have cost.
MAI-DxO does not rely on a single model answering at once. Microsoft says it works like a panel of virtual physicians that propose diagnoses, choose useful questions or tests, and challenge the emerging view over several rounds, weighing each test's expected value against its price. The orchestrator is model-agnostic; Microsoft and TechCrunch report it was tested with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, with the strongest results coming from OpenAI's GPT-5 and GPT-5.5.
According to Microsoft, MAI-DxO reached 83% accuracy on the NEJM cases, compared with 36% for generalist physicians who could use references and 22% for physicians working without them. On the emergency-department set it scored 80%, versus 47% for physicians. Medical Economics, reporting on the paper, gives the precise figures as 83.2% and 80.3% for the system against 36.1% and 46.8% for physicians with search and reference tools. Microsoft also says a cost-optimized configuration reached 81% on the NEJM cases at less than half the cost of comparable systems.
Independent coverage adds important context. TechCrunch notes that the study's data suggests much of the improvement comes from more capable underlying models, with the orchestration adding gains on top, and that the boost was largest on complex cases and with weaker models. On the emergency-department cases, TechCrunch reports little difference in accuracy or cost between GPT-5.5 with and without the orchestrator.
The limitations are substantial, and the authors list them. The comparison was against individual generalist physicians, not specialists or multidisciplinary teams, who would typically handle cases this complex. The benchmark was text-only, with no imaging, waveforms, or pathology slides; every case had a confirmed pathological diagnosis, so healthy and benign visits were excluded; and the study measured only diagnosis, not treatment decisions, patient safety, or outcomes. Medical Economics also reports that all authors are Microsoft employees who hold equity in the company, that eight are named inventors on a related Microsoft patent application, and that the study received no outside funding.
Microsoft says it is open-sourcing SDBench for select NEJM cases, including the benchmark, its gatekeeper and judge components, and cost-estimation code, so others can reproduce or challenge the findings. The company also announced a partnership with Beth Israel Deaconess Medical Center in Boston for a prospective study in which consenting patients will interact with the system directly while receiving standard care, with its recommendations compared against physicians' independent assessments. No timeline for clinical deployment has been announced.
DigiEditorial treats Microsoft AI's October 22 announcement as the primary source and corroborated the publication date, benchmark design, accuracy figures, and stated limitations against reporting from TechCrunch and Medical Economics. The accuracy claims describe performance in a controlled simulation and should not be read as evidence that the system is safe or effective in patient care.