Google ships MedGemma, open medical VLMs built on Gemma 3

Google has published MedGemma, a family of open medical vision-language models built on its Gemma 3 base, in a Nature Medicine paper dated October 6, 2026. The release spans 4B and 27B multimodal variants, a 27B text-only model, and MedSigLIP, a 400-million-parameter companion vision encoder.

On MedQA, the authors report an accuracy of 86.2%, which they describe as "the best performance for an open model" at release. Against the base Gemma 3, MedGemma reports gains of 15.5–18.1% on chest X-ray finding classification, 2.6–10% on medical image question answering, and 10.8% on agentic evaluations.

The models target a wide spread of modalities, including chest X-ray report generation, histopathology, dermatology, retinal imaging, electronic health record question answering, and clinical trial patient matching. The authors frame MedSigLIP as "a medically tuned vision encoder derived from SigLIP that powers the visual understanding capabilities of MedGemma," which teams can lift out and reuse on their own stacks.

The paper's practical pitch is aimed at teams working with modest clinical datasets. "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," the authors write, framing the result as "performance similar to or exceeding prior models such as Med-Gemini and larger frontier models" despite the smaller size. Two researchers we track posted the paper the same day it went live.