Google releases MedGemma, open medical VLMs on Gemma 3

Google's medical AI team has openly released MedGemma, a family of vision-language models built on Gemma 3 and tuned for healthcare tasks, in a paper published in Nature Medicine. The collection includes 4B and 27B multimodal variants, a 27B text-only sibling, and MedSigLIP, a 400M-parameter medical image encoder derived from SigLIP.

The headline number: 86.2% on MedQA, which the authors call the best open-model result at release. On out-of-distribution evaluations, the paper reports improvements of 2.6–10% in medical image question answering, 15.5–18.1% in chest X-ray finding classification, and 10.8% in agentic tasks over the base Gemma 3.

The abstract claims MedGemma "demonstrates advanced medical understanding and reasoning across images and text and multiple medical imaging domains, exceeding the performance of similarly sized generative models while maintaining the general capabilities of the Gemma base models."

The clinical evaluation worth stressing: MedGemma 27B's generated chest X-ray reports were rated clinically equivalent or superior to the originals in 81% of a 306-case evaluation drawn from MIMIC-CXR. The paper also finds that "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 of the researchers we follow in our Who's Who tracker have already shared it.