ResearchPublished All news from that day
Microsoft says MAI-DxO reached 83% accuracy on 304 NEJM cases, against 36% for physicians with resources.
Microsoft reports 83% accuracy against 36% for physicians with reference resources, in Microsoft's own evaluation of 304 NEJM cases.
Check our sources · 7 facts from 1 sourceMicrosoft AI published Building Medical Superintelligence on 22 October 2026, alongside a Science paper and a BMJ Digital Health and AI piece.
It describes MAI-DxO as a panel of virtual physicians that propose diagnoses, choose questions or tests, and challenge each other over several debate rounds while weighing each test's cost.
In Microsoft's evaluation on 304 complex New England Journal of Medicine cases, MAI-DxO reached 83% accuracy, against 22% for physicians without reference resources and 36% with them. On emergency cases from MIMIC-IV-ED it reached 80% against 47% for physicians.
Microsoft says the gains held when it repeated the evaluation with newer models including GPT-5.5. It says SDBench, the sequential diagnosis benchmark, is being open-sourced for select NEJM cases. Microsoft names a collaboration with the Mayo Clinic and a prospective study at Beth Israel Deaconess Medical Center.
Underlined sentences link to the facts we checked. Dotted words open a short explainer.
Fact 2 of 7 · from Microsoft AI, 22 Oct 2026
operates like a panel of virtual physicians
Fact 5 of 7 · from Microsoft AI, 22 Oct 2026
repeated the evaluation with newer models, including GPT-5.5, without changing the orchestration method
Fact 7 of 7 · from Microsoft AI, 22 Oct 2026
patients will interact with the system directly
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against this source.
1 Building Medical Superintelligence
Open the source archived copy-
Microsoft AI's post is dated 22 October 2026 and was published alongside a Science paper and a BMJ Digital Health and AI piece.
citeOur work is published today in Science.
-
Microsoft describes MAI-DxO as working like a panel of virtual physicians that propose diagnoses, choose useful questions or tests, and challenge each other over several debate rounds, weighing each test's likely value against its cost.
citeoperates like a panel of virtual physicians
-
Microsoft says that on 304 complex NEJM cases MAI-DxO reached 83% accuracy, against 22% for physicians without reference resources and 36% with them. the company says
citeOn complex NEJM cases, MAI-DxO achieved 83% diagnostic accuracy.
-
Microsoft says that on common emergency cases from MIMIC-IV-ED MAI-DxO reached 80% accuracy against 47% for physicians. the company says
citeMAI-DxO achieved 80% accuracy versus 47% for physicians.
-
Microsoft says the gains persisted when the evaluation was repeated with newer models including GPT-5.5, with the orchestration method unchanged. the company says
citerepeated the evaluation with newer models, including GPT-5.5, without changing the orchestration method
-
Microsoft says SDBench, including the benchmark, Gatekeeper, Judge and cost-estimation code, is being open-sourced for select NEJM cases.
citeWe are open-sourcing SDBench for select NEJM cases.
-
Microsoft names a collaboration with the Mayo Clinic and a prospective study with Beth Israel Deaconess Medical Center in which patients interact with the system directly.
citepatients will interact with the system directly
Every sentence above is checked against this source. Each fact has its own address you can cite.
Source 1 · Microsoft AI · 22 Oct 2026
Building Medical Superintelligence
Microsoft AI's post is dated 22 October 2026 and was published alongside a Science paper and a BMJ Digital Health and AI piece.
Our work is published today in Science.
Microsoft describes MAI-DxO as working like a panel of virtual physicians that propose diagnoses, choose useful questions or tests, and challenge each other over several debate rounds, weighing each test's likely value against its cost.
operates like a panel of virtual physicians
Microsoft says that on 304 complex NEJM cases MAI-DxO reached 83% accuracy, against 22% for physicians without reference resources and 36% with them. the company says
On complex NEJM cases, MAI-DxO achieved 83% diagnostic accuracy.
Show all 7 factsShow fewer facts
Microsoft says that on common emergency cases from MIMIC-IV-ED MAI-DxO reached 80% accuracy against 47% for physicians. the company says
MAI-DxO achieved 80% accuracy versus 47% for physicians.
Microsoft says the gains persisted when the evaluation was repeated with newer models including GPT-5.5, with the orchestration method unchanged. the company says
repeated the evaluation with newer models, including GPT-5.5, without changing the orchestration method
Microsoft says SDBench, including the benchmark, Gatekeeper, Judge and cost-estimation code, is being open-sourced for select NEJM cases.
We are open-sourcing SDBench for select NEJM cases.
Microsoft names a collaboration with the Mayo Clinic and a prospective study with Beth Israel Deaconess Medical Center in which patients interact with the system directly.
patients will interact with the system directly
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep