Two recent studies published in *Nature* show specialist health AI tools outpacing physicians in diagnostic and treatment decisions. Germany‑based Mira surpassed a panel of six doctors across eight conditions, while Google’s Amie generated more guideline‑aligned care plans than 21 primary‑care clinicians. The findings mark the first head‑to‑head evidence that narrow‑focus large language models can deliver superior medical advice.

Mira accessed electronic health‑record data for over 500 emergency cases, selecting from 85,000 possible actions. It achieved an 87.1% diagnostic accuracy, compared with the doctors’ 78.1% average. Amie, built on Google’s Gemini model, was tested on 100 multi‑visit scenarios derived from UK guidelines and matched physicians in reasoning while exceeding them on medication choices in complex cases.

Researchers stress the simulations were tightly controlled; patient chat inputs were cleaner than real‑world ED speech and both systems occasionally suggested suboptimal care. Academics from Oxford and Edinburgh praised the rigor but warned the results do not yet translate to clinical settings. Investors will watch how these prototypes evolve, as proven superiority could reshape diagnostic software markets and trigger health‑tech partnerships in the near term.