> ## Content Index > Fetch the complete content index at: https://theainewsfiles.com/llms.txt > Use this file to discover other available public pages before exploring further. # Microsoft says its AI diagnostic tool beats doctors: what the new figures say - URL: https://theainewsfiles.com/microsoft-diagnostic-orchestrator-ai-vs-doctors/ - Published: 2026-10-25T17:02:38.000Z - Updated: 2026-10-25T17:02:38.000Z - Description: The Verge reports Microsoft saying its Diagnostic Orchestrator can diagnose up to four times more accurately than human doctors. Here are the new figures and how they compare with Microsoft's 2025 preprint. - Author: J E Cowie - Tags: News Microsoft claims to have developed an AI-powered tool that can diagnose conditions up to four times more accurately than human doctors, The Verge reports. Mustafa Suleyman, who runs Microsoft AI, told The Verge the moment "is finally here". This is a company's claim about its own system, tested on past cases. Read it that way. ![Map showing where Beth Israel Deaconess Medical Center is](https://theainewsfiles.com/content/images/2026/10/map_beth-israel-deaconess-medical-center.jpg) Beth Israel Deaconess Medical Center, Massachusetts, United States ## What Microsoft is claiming The Verge's Robert Hart reported on 22 October 2026 that the findings were published on Thursday in *Science*. The tool is called the Diagnostic Orchestrator. According to The Verge, it works like a panel of medical experts: several AI agents reason through symptoms and medical histories, decide which tests to order and arrive at a diagnosis. The Verge reports two sets of results: - **Published case reports.** On cases from hundreds of *New England Journal of Medicine* reports, the system reached a diagnostic accuracy of around 83 per cent. The Verge gives roughly 36 per cent for generalist physicians using references and 22 per cent for those working without them. - **Emergency department cases.** On cases drawn from a large database of patient admissions at Beth Israel Deaconess Medical Center in Boston, the system scored 80 per cent against an average of 47 per cent for human physicians. The Verge also reports that the system tended to reach diagnoses at a lower estimated cost than the human physicians. It says Microsoft compiled the cases into its own benchmark, SDBench, which the company plans to open-source. Microsoft's earlier blog post said only that it was exploring the potential of launching SDBench as a public benchmark. ![The exterior of Beth Israel Deaconess Medical Center in Boston](https://theainewsfiles.com/content/images/2026/10/shot2_the-exterior-of-beth-israel-deaconess-me.jpg) Beth Israel Deaconess Medical Center in Boston. ## How this compares with the 2025 preprint Microsoft's earlier account of this work is a preprint on arXiv, "Sequential Diagnosis with Language Models", first submitted on 27 June 2025 and revised on 2 July 2025, according to the arXiv listing. Microsoft's own blog post, written by Dominic King and Harsha Nori, said the company was in the process of submitting the work for external peer review. The preprint turned 304 difficult NEJM case reports into step-by-step diagnostic encounters. A doctor or an AI starts with a short case summary and has to ask for further details. The preprint's headline numbers were: - paired with OpenAI's o3 model, the orchestrator reached 80 per cent accuracy against a 20 per cent average for generalist physicians; - configured for maximum accuracy, it reached 85.5 per cent; - it cut diagnostic costs by 20 per cent compared with physicians. The blog post added that the 21 physicians in the comparison had 5 to 20 years of clinical experience. Several of the figures The Verge reports differ from the preprint's. The Verge gives around 83 per cent where the preprint gave 80 per cent for the o3 pairing and 85.5 per cent for the maximum-accuracy configuration. It gives two physician figures (36 and 22 per cent) where the preprint gave a single 20 per cent average. It says the system is powered by OpenAI's GPT-5, where the preprint paired it with o3\. The preprint's abstract describes its benchmark only in terms of NEJM case reports, and the earlier blog post does not mention emergency department cases from Beth Israel Deaconess at all. The figures above are The Verge's account. My reading is that the four-fold comparison is the company's headline (its preprint called the result "four times higher" than physicians), and that how large the gap looks depends on which physician baseline a reader picks. That is a point about presentation, not an allegation about the research. ## What the system has not been shown to do The Verge says Microsoft itself acknowledges a long road ahead. As of The Verge's report on 22 October, the system had been tested only on past medical cases, and it remained unclear how well it would perform in real clinical settings. "It needs clinical validation," Suleyman told The Verge. The Verge reports that the paper does not include newer models such as GPT-6 Astra, Meta's Muse Wonder Full, Google's Gemini 4 Argon or Anthropic's Opus 5.5\. It adds that the orchestrator was first developed when GPT-4.1 was the most advanced model available. The researchers say later testing suggests that "improvements in model capability do not necessarily diminish the benefits" of having multiple agents work together, according to The Verge. Microsoft's earlier blog post described the orchestrator and SDBench as research demonstrations. It said they were not available as public tools and the work was not approved for clinical use. It also listed limitations: more testing would be needed on common, everyday presentations, and the physicians worked without access to colleagues, textbooks or generative AI, which the blog said can form part of normal clinical practice. ## What comes next, according to Microsoft The Verge reports that King, a coauthor of the study whom The Verge describes as Microsoft's AI health lead, said Microsoft plans to begin testing the system in hospital settings with Beth Israel Deaconess Medical Center. Neither King nor Suleyman would commit to a firm timeline, though King told The Verge it will likely "get underway in the next couple of months". The Verge adds that, as of its 22 October report, Microsoft had yet to say when or how the technology might become available commercially. Suleyman was more bullish about the wider picture. "Next year we're going to see extremely impressive deployments of intelligent systems in production," he told The Verge. He also said human clinicians should and will always have a central role in patient care. The work is part of what Microsoft calls "medical superintelligence", according to The Verge, which also says Microsoft frames diagnosis as one part of a broader health-AI vision, set out in a blog post published Thursday and an article in the BMJ Digital Health & AI journal. ## Where this leaves things Announced: a tool that Microsoft says beats physicians on diagnosis, reported, according to The Verge, in the prestigious journal *Science*. Shipped: nothing yet, by The Verge's 22 October account, which says Microsoft had not said when or how the technology might become available. Independently verified: that depends on testing outside Microsoft's own benchmark. The hospital testing King described with Beth Israel Deaconess, if it goes ahead, is what would show whether the results survive contact with real patients. ## Sources Hart, R. (2026) 'Microsoft says its AI tool is better than doctors at diagnosing conditions', The Verge, 22 October 2026, available at: [https://www.theverge.com/ai-artificial-intelligence/1016690/suleyman-microsoft-diagnostic-orchestrator-science-medical-superintelligence](https://www.theverge.com/ai-artificial-intelligence/1016690/suleyman-microsoft-diagnostic-orchestrator-science-medical-superintelligence?ref=theainewsfiles.com) (Accessed: 25 October 2026). King, D. and Nori, H. (2025) 'El camino hacia la superinteligencia médica', Source LATAM, Microsoft, 30 June 2025, available at: [https://news.microsoft.com/source/latam/noticias-de-microsoft/el-camino-hacia-la-superinteligencia-medica](https://news.microsoft.com/source/latam/noticias-de-microsoft/el-camino-hacia-la-superinteligencia-medica?ref=theainewsfiles.com) (Accessed: 25 October 2026). Nori, H., Daswani, M., Kelly, C., Lundberg, S., Ribeiro, M.T., Wilson, M., Liu, X., Sounderajah, V., Carlson, J., Lungren, M.P., Gross, B., Hames, P., Suleyman, M., King, D. and Horvitz, E. (2025) 'Sequential Diagnosis with Language Models', arXiv:2506.22405, submitted 27 June 2025, revised 2 July 2025, available at: [https://arxiv.org/abs/2506.22405](https://arxiv.org/abs/2506.22405?ref=theainewsfiles.com) (Accessed: 25 October 2026). ## Picture credits - Top picture: Mustafa Suleyman, chief executive of Microsoft AI. Mustafa Suleyman, photo by Christopher Wilson, [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/?ref=theainewsfiles.com) (via [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Mustafa%5FSuleyman%5Fphoto%5F%28cropped%29.jpg?ref=theainewsfiles.com)) - Beth Israel Deaconess Medical Center in Boston. Beth Israel Deaconess Medical Center, Boston, photo by Ajay Suresh, [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/?ref=theainewsfiles.com) (via [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Beth%5FIsrael%5FDeaconess%5FMedical%5FCenter%5F%28BIDMC%29%5F%2854953356537%29.jpg?ref=theainewsfiles.com)) - Map: drawn from [Natural Earth](https://www.naturalearthdata.com/?ref=theainewsfiles.com) (public domain) and [Wikidata](https://www.wikidata.org/wiki/Q4897536?ref=theainewsfiles.com).