Google's MedGemma posts gains on chest X-ray and MedQA tasks

Google has published MedGemma, a family of open medical vision-language models built on its Gemma 3 base, in Nature Medicine. The release includes 4B and 27B parameter multimodal variants, a 27B text-only version, and MedSigLIP, a 400M-parameter medical vision encoder.

Against their Gemma base models, the authors report "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 evaluations." On MedQA the paper reports 86.2% accuracy, described as the best open model performance at release, and in chest X-ray report generation 81% of reports rated as producing equal or superior clinical decisions. Two researchers we follow had circulated the paper by the day it landed.

MedSigLIP, the vision encoder behind MedGemma, is described as achieving "performance comparable to or better than that of many specialized medical image encoders."