adtestbench Higgsfield

AI news · Microsoft · health AI

Microsoft says its AI beat generalist doctors on hard cases

In a study published in Science, MAI-DxO reached the right diagnosis in 83 per cent of complex cases, against 36 per cent for doctors using references.

By Alexandre S. , 06:30 UTC

Microsoft AI says its diagnostic system, MAI-DxO, solved hard medical cases far more often than generalist doctors, in a study published in Science on Thursday 22 October 2026, according to the company. Chief executive Mustafa Suleyman called it a step towards “medical superintelligence”, TechCrunch reports.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a software window standing on the horizon.
Drawn by adtestbench from “Building medical superintelligence”,

MAI-DxO seats several AI agents around one simulated patient, like a hospital case conference. They ask questions, order tests and challenge each other’s diagnosis over several rounds, pricing each test as they go. On the hardest New England Journal of Medicine cases it was right 83 per cent of the time. Generalist doctors got 22 per cent right alone and 36 per cent with reference materials.

Specialists and physician teams, who usually take cases this hard, were left out of the comparison, TechCrunch notes. Much of the gain comes from the underlying model: MAI-DxO did best on OpenAI’s GPT-5 and GPT-5.5, and on GPT-5 the panel set-up added about four points of accuracy while cutting the estimated cost per case by about $3,000.

Microsoft’s 83 per cent was measured against generalists on simulated patients, and most of it came from the model underneath, so it says nothing yet about the specialist teams who usually take such cases.