Google publishes MedGemma, open medical vision-language models

Google has released MedGemma, a family of open medical vision-language models built on Gemma 3, with reported gains of 15.5–18.1% on chest X-ray finding classification over its base. The paper appears in Nature Medicine.

The release spans MedGemma 4B and 27B multimodal variants plus a 27B text-only model, alongside MedSigLIP, a 400-million-parameter companion vision encoder the authors describe as "a medically tuned vision encoder derived from SigLIP that powers the visual understanding capabilities of MedGemma." Beyond chest X-rays, the paper reports 2.6–10% improvements on medical image question answering and 10.8% on agentic evaluations compared with base Gemma 3.

The data-efficiency pitch is where the paper leans hardest: "Fine-tuning MedGemma can be more effective than fine-tuning the base Gemma 3 model for medical tasks, particularly in the setting of limited training data."

Two researchers we track in our Who's Who directory had already circulated the Nature link.

"Artificial intelligence has high potential for impact in healthcare applications, but its training and deployment are challenging due to diverse data, a complex spectrum of possible tasks and important privacy needs," the abstract opens, framing the open-weights release as a tractability and privacy play. The abstract reports headline deltas rather than raw per-task accuracy figures, and stays silent on commercial licensing terms for clinical deployment.