Project AURIONAI Use-Cases Research & Innovation
Google
multimodal LLM

MedGemma 4B

4B multimodal medical model built on Gemma 3 4B, used for radiology report generation and medical QA. Dermatology images among its inputs is unverified in the notes.

GoogleGoogleUSAMedGemmaFamily
Figure from the MedGemma 4B paper: Example of histopathology open-ended question answering with MedGemma 4B.
Levelmodel
Classmultimodal LLM
Parameters4B
Contextnot in the notes
Releasednot in the notes
Opennessopen weights restricted
Runs onnot in the notes
Statusnot in the notes

Benchmark Results

vendor-reported
CheXpertdomain
Out of distributionvendor-reported
48.1
04590

Score in macro F1. Higher is better.

Jul 2025 arxiv.org
MedQAdomain
vendor-reported
64.4
050100

Score in %. Higher is better.

Jul 2025 arxiv.org, huggingface.co
MIMIC-CXRdomain
Med-Gemini test splitvendor-reported
88.9
04590

Score in macro F1. Higher is better.

Jul 2025 arxiv.org
RadGraph F1domain
Fine-tunedvendor-reported
30.3
02040

Score in F1. Higher is better.

Jul 2025 arxiv.org
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

finetunesuccessor
Gemma 3 4B to MedGemma 4B: finetune, high confidenceMedGemma 4B to MedGemma 1.5 4B: successor, high confidenceGemma 3 4BGemma 3 4BGoogle, Mar 2025MedGemma 4BMedGemma 4BGoogleMedGemma 1.5 4BMedGemma 1.5 4BGoogle, Apr 2026
GoogleGemma 3 4B is the parent (finetune)high confidencemodel card
GoogleMedGemma 1.5 4B is a child (successor)high confidencenotes
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 3 models. Confidence and evidence are the notes' own.

Industry Use Cases

Healthcare
Clinical LLM (open weights)DermatologyRadiology report generation

R&D Topics

Domain adaptation (vertical models)produced byMedical vision-language modelsexemplar

Identity

Developer
Google
Family
MedGemma
Released
– sources
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
not in the notes
Total parameters
4B sources
Active parameters
–
Checkpoint parameters
–
Context length
not in the notes
Input
text, image
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open weights restricted
Licence
Health AI Developer Foundations terms of use sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
not in the notes
Footprint
On premises or in the cloud (open weights).
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Sources

  1. MedGemma Technical Report arxiv.org https://arxiv.org/html/2507.05201v2
  2. google/medgemma-27b-text-it , Hugging Face huggingface.co https://huggingface.co/google/medgemma-27b-text-it

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MedGemma 4B", https://aurion.semai.id/models/google-medgemma-4b/, records of October 2026.