Back to Models
DeepMindProprietary

MedGemma 4B Instruct

MedGemma 4B Instruct, developed by MedMind, is a foundation model specialized in the medical field and presented in a chat format. It has a parameter scale of approximately 40 billion and supports a long context window of up to 128K.

Parameters

40.0B

Context Window

128K

License

https://creativecommons.org/licenses/by-sa/4.0/

Release Date

2025-07-09

API Pricing

API pricing for this model is not yet available

Strengths

  • Sufficient 4B parameter scale
  • Vast 128K context processing
  • Medical-specialized chat optimization

Weaknesses

  • Restrictions from closed license
  • High resource consumption due to size
  • Unclear adaptability to general tasks

Use Cases

  • Medical information chatbot development
  • Analysis/summary of lengthy medical papers
  • Generation of specialized medical dialogues

Deep Analysis

Parameters

4B

Multimodal variant, can run on mobile hardware

MedQA

64.4%

Among best very small (<8B) open models

Chest X-ray Accuracy

81% clinically sufficient

US board-certified radiologist judged reports comparable to original

Base Model

Gemma 3

Medically optimized variant with SigLIP image encoder

Release Date

July 9, 2025

Part of HAI-DEF collection

Deployment

Mobile-capable

Can be adapted to run on mobile hardware

Image Encoder

MedSigLIP

Pre-trained on chest X-rays, dermatology, ophthalmology, histopathology

License

HAI-DEF Terms of Use

Health AI Developer Foundations license

Strengths

  • Runs on mobile hardware - extremely accessible medical AI
  • 64.4% MedQA is excellent for a 4B model
  • 81% clinically sufficient chest X-ray reports after radiologist review
  • Multimodal: handles both medical text and images
  • Retains general Gemma capabilities for mixed medical/non-medical tasks
  • Competitive with task-specific state-of-the-art on medical image classification

Weaknesses

  • Significantly lower MedQA than 27B variant (64.4% vs 87.7%)
  • HAI-DEF license more restrictive than Apache 2.0
  • Requires fine-tuning for best performance on specific tasks
  • Limited to text and image modalities
  • Not suitable for complex multi-step medical reasoning

Competitor Comparison

ModelArenaSWEGPQAPrice
MedGemma 4BN/AN/A64.4% (MedQA)Free (open weights)
MedGemma 27BN/AN/A87.7% (MedQA)Free (open weights)
BioMistral 7BN/AN/A~60% (MedQA)Free (open weights)
Gemma 3 4B (base)~1200N/A~45% (MedQA est)Free (open weights)

MedGemma 4B is Google's compact medical AI model that can run on mobile hardware. With 64.4% MedQA (among best under 8B) and 81% clinically sufficient chest X-ray reports, it provides accessible medical AI for resource-constrained deployments. Part of the Health AI Developer Foundations collection.

Analysis generated: 2026-05-24