At the recently concluded 2025 I/O Developer Conference, Google announced the open-sourcing of its brand-new medical AI model — MedGemma. This model, based on the Gemma3 architecture, is specifically designed for the healthcare sector and features strong multi-modal image and text understanding capabilities, aiming to enhance the efficiency of medical diagnosis and treatment.
MedGemma offers two configuration options: a 4B-parameter model and a 27B-parameter model. The 4B-parameter model is primarily used for classifying and interpreting medical images, capable of generating detailed diagnostic reports or answering questions related to images; the 27B-parameter model, on the other hand, focuses on processing clinical texts, making it particularly suitable for patient triage and decision support. This means that doctors can quickly obtain valuable information about patients' conditions, allowing them to more effectively develop treatment plans.
Image source note: Image generated by AI, image authorization service provider Midjourney
Among these, the image encoder of the 4B model uses SigLIP technology, with pre-trained data covering images from multiple fields such as chest X-rays, dermatology, ophthalmology, and histopathology, enabling it to perform exceptionally well in handling complex medical images. By contrast, the 27B model, with its 27 billion parameters, focuses on deeply understanding medical texts, providing support for clinical reasoning and helping medical staff make decisions more quickly.
Even more flexibly, developers can choose to run these models locally for experiments or deploy them as HTTPS endpoints via Google Cloud's Vertex AI platform for large-scale applications. Google also provides abundant resources like Colab notebooks to help developers fine-tune and integrate the models.
In addition, Google encourages developers to validate and fine-tune the models for specific use cases and provides relevant guidance and tools to support efficient parameter fine-tuning using methods such as prompt engineering, context learning, and LoRA during the adaptation process. This makes MedGemma not only an advanced AI tool but also an important assistant for developers and medical personnel to improve their daily work efficiency.
The launch of MedGemma marks another major advancement in the field of medical AI, bringing endless possibilities for future medical practice.