Google is making a significant move in healthcare AI by releasing its latest MedGemma and MedSigLIP models as open-source tools. Instead of keeping these advanced systems behind expensive paywalls, Google is letting hospitals, researchers, and developers download, modify, and run them independently.
The MedGemma 27B Multimodal model stands out for its ability to process both medical text and images, such as chest X-rays and pathology slides, allowing it to analyze complex patient information like a real doctor. In benchmark tests, the model scored an impressive 87.7% on MedQA, matching the performance of much larger systems but with far lower operating costs. This can be a game-changer for healthcare providers with limited budgets.
A smaller version, MedGemma 4B, still performs strongly, achieving 64.4% on the same benchmarks. Importantly, expert radiologists found its chest X-ray reports accurate enough to help guide patient care in most cases.
Alongside these, Google released MedSigLIP, a lighter model focused on medical images, trained on everything from skin photos to tissue samples. Despite its compact size, MedSigLIP excels at spotting medically relevant details and can link images to similar cases based on clinical context.
Healthcare organizations are already testing these models. In the US, DeepHealth uses MedSigLIP to help radiologists catch overlooked problems in chest X-rays. In Taiwan, MedGemma has shown high accuracy with traditional Chinese medical texts, while Indian startup Tap Health reports that MedGemma provides reliable, context-aware medical answers.
Open-sourcing these AI models is crucial for healthcare’s unique needs. Hospitals can keep patient data secure, adapt the models for specific tasks, and ensure stable performance over time. Google stresses, however, that these tools should assist—not replace, doctors, as clinical oversight and human judgment remain vital.
The accessibility of these models, even on basic hardware, opens opportunities for smaller hospitals and developing regions to use advanced AI. By amplifying human expertise and supporting overworked medical staff, Google’s MedGemma and MedSigLIP could help make healthcare more efficient and accessible worldwide.