Google AI releases Medgemma: an open suite of models trained on medical text and image understanding

At Google I/O 2025, Google launched Medgemma, an open model suite designed for multimodal medical text and image understanding. Built on the Gemma 3 architecture, Medgemma is designed to provide developers with a strong foundation to create medical applications that require integrated analysis of medical images and text data.

Model variants and architectures

Medgemma is available in two configurations:

• Medgemma 4b: A 4 billion parameter multi-model capable of processing medical images and text. It uses a pre-trained siglip image encoder with pre-identified medical data sets, including chest X-rays, dermatological images, ophthalmic images and histopathology slides. The language model components are trained in a variety of medical data to promote a comprehensive understanding.
• Medgemma 27b: A 27 billion parameter text model optimized for tasks requiring in-depth medical text understanding and clinical reasoning. This variant is specifically tailored by guidance and is designed for applications requiring advanced text analysis.

Deployment and accessibility

Developers can access Medgemma models by embracing faces, but agree to the Healthy AI Developer Basic Terms of Use. These models can be run locally for experimentation, or through Google Cloud’s Pertex AI for production-grade applications. Google provides resources including COLAB notebooks to facilitate fine-tuning and integration into a variety of workflows.

Applications and Use Cases

Medgemma is the basic model for several healthcare-related applications:

• Medical image classification: The pre-training of the 4B model makes it suitable for classification of various medical images, such as radiological scans and dermatological images.
• Medical image interpretation: It can generate reports or answer questions related to medical images and help with the diagnostic process.
• Clinical text analysis: The 27B model performed well in understanding and summarizing clinical notes, supporting patient triage and decision support tasks.

Adaptation and fine-tuning

While Medgemma provides strong baseline performance, developers are encouraged to validate and tweak models for their specific use cases. Technologies such as timely engineering, in-house cultural learning and parameter fine-tuning methods such as Laura can be used to enhance performance. Google provides guidance and tools to support these adaptation processes.

in conclusion

Medgemma represents an important step in providing accessible open source tools for medical AI development. By combining multimodal functionality with scalability and adaptability, it provides an invaluable resource for applications designed to integrate medical image and text analysis.

View models on hug faces and project pages. All credits for this study are to the researchers on the project. Also, please feel free to follow us twitter And don’t forget to join us 95k+ ml reddit And subscribe Our newsletter.