Medical imaging prototypes
Adapt a multimodal model for radiology, dermatology, pathology, ophthalmology, CT, MRI, or longitudinal imaging workflows.
Independent tool overview
MedGemma at a glance
MedGemma is Google's collection of free open-weight Gemma 3 models for developers building and validating medical text, imaging, and clinical-reasoning applications.
Visit the official MedGemma site ↗
- Current release
- MedGemma 1.5 4B multimodal
- Other variants
- MedGemma 1 in 4B and 27B text or multimodal forms
- Context
- At least 128K tokens
- License
- HAI-DEF terms; companion code is Apache 2.0
- Weights
- Free for research and commercial use
- Clinical status
- Developer model; not clinical-grade without validation and adaptation
Overview
What MedGemma is
MedGemma is a collection of Google Health AI Developer Foundations models adapted from Gemma 3 for medical text and image comprehension. It is a developer building block, not a patient-facing chatbot or a ready-to-deploy clinical product. Teams download the weights, deploy them in their own environment or cloud account, and adapt the models to a narrowly defined healthcare use case.
The current collection includes MedGemma 1.5 as a 4B multimodal instruction-tuned model, plus MedGemma 1 variants at 4B and 27B. MedGemma 1.5 improves medical reasoning, medical-record interpretation, lab-report extraction, CT and MRI volume interpretation, whole-slide pathology, longitudinal chest X-rays, and anatomical localization. Google continues to recommend the larger 27B model for more complex text-heavy applications.
The weights are free for research and commercial use under Google's Health AI Developer Foundations terms, but they are open-weight rather than open-source. MedGemma is not clinical-grade out of the box. Developers are responsible for evaluation, adaptation, patient-data protection, regulatory compliance, and independent verification before any real-world medical use.
Use cases
Who MedGemma is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Clinical text and EHR research
Build and evaluate systems for summarization, structured extraction, patient intake, triage support, or clinical reasoning.
Private or offline deployment
Run open weights on local, edge, air-gapped, or controlled cloud infrastructure when data sovereignty matters.
Healthcare model adaptation
Fine-tune, prompt, or orchestrate a healthcare-specific base model against an organization's own data and validated task.
Capabilities
Core MedGemma features
MedGemma 1.5 4B multimodal
A compute-efficient text-and-image model with improvements across medical reasoning, records, documents, and multiple medical imaging modalities.
MedGemma 1 27B variants
Larger text-only and multimodal options for workloads where stronger baseline medical text performance justifies higher compute requirements.
High-dimensional imaging support
MedGemma 1.5 expands adaptation targets to CT and MRI volumes, whole-slide histopathology, longitudinal chest X-rays, and anatomical localization.
Medical document understanding
Extract structured information from lab reports and interpret text-based EHR data as part of a validated application workflow.
Open-weight deployment
Download the weights from Hugging Face and run them locally, on private infrastructure, or through a cloud deployment selected by the developer.
Fine-tuning and orchestration
Adapt the models with prompt engineering, in-context examples, supervised fine-tuning, retrieval, or a larger agentic system.
Google Cloud integration
Use Model Garden presets, managed batch jobs, endpoints, GKE, DICOM and FHIR integrations, and Google Cloud governance controls.
Long context
The model cards document support for at least 128K tokens, enabling long medical records and document collections subject to deployment memory.
Process
How the MedGemma workflow works
Step 1
Define a narrow intended use
Specify the exact input, output, users, clinical setting, risks, and success criteria before choosing a model.
Step 2
Select and deploy a variant
Use MedGemma 1.5 4B for efficient multimodal work or evaluate a 27B MedGemma 1 model for more demanding text and reasoning tasks.
Step 3
Adapt on relevant data
Apply prompt engineering, retrieval, fine-tuning, or orchestration using representative, appropriately governed medical data.
Step 4
Validate before real-world use
Measure performance, safety, subgroup behavior, workflow fit, privacy, and regulatory requirements with qualified clinical and technical reviewers.
Cost
MedGemma pricing and free plan
Google provides MedGemma weights at no charge for research and commercial use under the HAI-DEF terms. There is no free hosted MedGemma application: users pay for their own local hardware or usage-based cloud compute, storage, networking, and operations.
MedGemma 1.5 4B weights
Free
The current compute-efficient multimodal model, distributed under the HAI-DEF terms.
- Research and commercial use permitted under the license
- Hugging Face access requires accepting the terms
- Approximate static weight footprint: 6.4 GB BF16, 4.4 GB 8-bit, or 3.4 GB 4-bit
- Hosting and adaptation costs are separate
MedGemma 1 27B weights
Free
Larger model options for stronger baseline text performance and selected multimodal workloads.
- Text-only and multimodal variants are available
- Approximate static weight footprint: 46.4 GB BF16, 29.1 GB 8-bit, or 21 GB 4-bit
- Runtime and KV-cache memory are additional
- Hosting and adaptation costs are separate
Self-hosted deployment
Infrastructure cost
Run MedGemma on local, private-cloud, edge, or air-gapped infrastructure.
- No per-token fee paid to Google for the weights
- Hardware, electricity, engineering, monitoring, and security costs apply
- Provides control over where patient data is processed
- Requires the team to operate and validate the service
Managed cloud deployment
Usage-based
Deploy through Google Cloud or another hosting provider and pay for the underlying infrastructure.
- Batch jobs can avoid always-on endpoint costs
- Online endpoints can scale for low-latency applications
- Pricing depends on accelerator, region, uptime, storage, and traffic
- Use the selected provider's calculator for an estimate
Pricing checked . Check current pricing at the source ↗
Assessment
MedGemma strengths and limitations
Where it stands out
- Healthcare-specialized text and image models in compute-efficient and larger sizes
- Free downloadable weights support private, offline, edge, and multi-cloud deployments
- MedGemma 1.5 broadens support for high-dimensional and longitudinal imaging
- Can be customized deeply with prompts, retrieval, fine-tuning, and orchestration
- Google publishes model cards, benchmarks, tutorials, and cloud deployment paths
What to consider
- MedGemma is not a standalone application and has no hosted chat experience for end users
- Google says the baseline model is not clinical-grade and requires use-case-specific validation and likely adaptation
- MedGemma 1.5 has not been evaluated or optimized for multi-turn applications
- Model outputs are preliminary and require independent verification and clinical correlation
- The weights use HAI-DEF terms rather than an open-source software license, and clinical or commercial distribution may create regulatory obligations
Compare
MedGemma alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Content Creator
Gemini
A managed general-purpose Gemini experience when self-hosting and deep healthcare-specific model customization are unnecessary.
Explore Gemini →Healthcare
ChatGPT for Clinicians
A clinician-facing product for verified U.S. doctors who need a managed workflow rather than downloadable developer weights.
Explore ChatGPT for Clinicians →Healthcare
CompliantChatGPT
A managed healthcare-oriented assistant option for organizations prioritizing an application experience over model infrastructure.
Explore CompliantChatGPT →Questions
MedGemma FAQs
What is MedGemma?
MedGemma is Google's collection of open-weight Gemma 3 variants trained for medical text and image comprehension. It is designed as a developer foundation for healthcare AI applications.
Is MedGemma free?
The model weights are free for research and commercial use under the Health AI Developer Foundations terms. You still pay for hardware, hosting, adaptation, security, and operations.
Is MedGemma open source?
No. Google describes HAI-DEF models as open-weight, not open-source. The model weights use the HAI-DEF terms, while accompanying repository code is licensed under Apache 2.0.
Can MedGemma diagnose patients?
MedGemma is not clinical-grade out of the box and should not be used as an unvalidated diagnostic system. Any clinical application requires meaningful adaptation, independent verification, qualified oversight, and applicable regulatory authorization.
What is the difference between MedGemma 1.5 4B and MedGemma 1 27B?
MedGemma 1.5 4B is the current compute-efficient multimodal release with expanded medical imaging, records, and document capabilities. MedGemma 1 27B requires much more memory but is generally the stronger baseline for complex text-heavy tasks.
Can MedGemma run locally?
Yes. The weights can run on local or private infrastructure. Google's approximate static weight footprints range from 3.4 to 6.4 GB for 4B and 21 to 46.4 GB for 27B depending on precision, before runtime and context-cache overhead.
Does Google log patient data when MedGemma runs on Google Cloud?
Google's HAI-DEF FAQ says MedGemma deployed on Google Cloud processes data within the customer's project and that Google does not access or log patient information or use it to train foundation models. Organizations still remain responsible for configuration, privacy, BAAs, and compliance.
Bottom line
Our MedGemma verdict
MedGemma is a compelling foundation for teams that have the medical expertise, data governance, infrastructure, and evaluation program required to build a healthcare AI system responsibly. MedGemma 1.5 makes advanced imaging and record tasks more accessible in a 4B model, while the 27B options remain useful for harder text workloads. It is not a shortcut to a deployable clinical product; the real work is adaptation, validation, oversight, and compliance.
Visit MedGemma website ↗Similar Healthcare Tools
View all Healthcare AI tools →
Chai-2
Chai-2 - an AI model capable of creating functional antibodies for drug development
Health in ChatGPT
ChatGPT Health - Dedicated, private experience for health conversations within ChatGPT, enabling connecting medical data and wellness apps
Earkick
Personal AI for mental health tracking and support.
MealPractice
Unleash your culinary creativity with our recipe generator.
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.