MedGemma Technical Report

By Andrew B. Sellergren, Sahar Kazemzadeh, Tiam Jaroensri, A. Kiraly, Madeleine Traverse, Timo Kohlberger, Shawn Xu, Fayaz Jamil, Cían Hughes, Charles Lau and 71 more. 2025

365 citations recorded by the source corpus.

Artificial intelligence (AI) has significant potential in healthcare applications, but its training and deployment faces challenges due to healthcare's diverse data, complex tasks, and the need to preserve privacy. Foundation models that perform well on medical tasks and require less task-specific tuning data are critical to accelerate the development of healthcare AI applications. We introduce MedGemma, a collection of medical vision-language foundation models based on Gemma 3 4B and 27B. MedGemma demonstrates advanced medical understanding and reasoning on images and text, significantly exceeding the performance of similar-sized generative models and approaching the performance of task-specific models, while maintaining the general capabilities of the Gemma 3 base models. For out-of-distribution tasks, MedGemma achieves 2.6-10% improvement on medical multimodal question answering, 15.5-18.1% improvement on chest X-ray finding classification, and 10.8% improvement on agentic evaluations compared to the base models. Fine-tuning MedGemma further improves performance in subdomains, reducing errors in electronic health record information retrieval by 50% and reaching comparable performance to existing specialized state-of-the-art methods for pneumothorax classification and histopathology patch classification. We additionally introduce MedSigLIP, a medically-tuned vision encoder derived from SigLIP. MedSigLIP powers the visual understanding capabilities of MedGemma and as an encoder achieves comparable or better performance than specialized medical image encoders. Taken together, the MedGemma collection provides a strong foundation of medical image and text capabilities, with potential to significantly accelerate medical research and development of downstream applications. The MedGemma collection, including tutorials and model weights, can be found at https://goo.gle/medgemma.

Subfields: Computer Vision and Pattern Recognition, Health Informatics, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/d23432504ce8755dd9933b93fbffd66bd462c9c9

Citations in the corpus

• Large Language Models Predict Functional Outcomes after Acute Ischemic Stroke (2026, unknown): “1-8B is a general-purpose model pretrained on vast internet data 28 , whereas MedGemma-4B is a clinical model pretrained on medical text 29 .”
• Domain Fine-Tuning vs. Retrieval-Augmented Generation for Medical Multiple-Choice Question Answering: A Controlled Comparison at the 4B-Parameter Scale (2026, unknown): “MedGemma continues this trend in the open-weight regime by adapting Gemma 3 to medical text and images [10].”
• Federation of Agents: A Semantics-Aware Communication Fabric for Large-Scale Agentic AI (2025, unknown): “13 , a 13x relative improvement over the best single agent baseline (Medgemma [51]) and a 6.5x improvement over the uncoordinated ensemble.”
• RELATE: Relation Extraction in Biomedical Abstracts with LLMs and Ontology Constraints (2025, unknown): “MedGemma (27B) is Google’s medical language model (Sellergren et al., 2025), trained exclusively on medical text and optimized for efficient inference through architectural improvements in performance and computational efficiency.”