MAI-DxO
Microsoft's orchestration layer that runs LLM agents as a mock physician panel, scoring 83.2% on 304 NEJM cases in a Science paper.
This entry is marked for review. Check the original sources before relying on its details.
Overview
MAI-DxO is a Microsoft harness that coordinates multiple AI agents as a simulated panel of physicians to work through complex diagnostic cases. In a paper published in Science and covered on October 22, 2026, it scored 83.2% on 304 NEJM teaching cases when paired with GPT-5, against 36.1% for generalist physicians using reference resources. It raised GPT-5.5 from about 72% to nearly 87% and Gemini 2.5 Flash from 46.4% to nearly 69%; estimated cost per case was $2,241-$5,431 versus $3,049-$4,725 for physician workups. Limits: text-only cases, no specialists or care teams in the comparison, and much of the gain comes from the underlying model. It is being evaluated with Beth Israel Deaconess Medical Center and is not a public product.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Orchestrates multiple agents as a mock physician panel
- Model-agnostic: tested with OpenAI, Google, Anthropic, Meta and DeepSeek models
- 83.2% on 304 NEJM cases with GPT-5 versus 36.1% for generalist physicians
- Estimated $2,241-$5,431 per case
- Under evaluation with Beth Israel Deaconess Medical Center
- Microsoft plans a healthcare foundation model with Mayo Clinic targeted for 2027
Use cases
- Decision support for complex diagnostic workups (research stage)
- Benchmarking frontier models on sequential diagnosis tasks
Pricing
unknown; not commercially available.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 22 Oct 2026
- Record updated
- 28 Oct 2026
- Record last checked
- 28 Oct 2026