MedGemma
Open medical text and vision models built on Gemma 3.
About
MedGemma is Google's open medical model family, released at Google I/O 2025: a 4B multimodal model that reads medical images and text, plus a 27B text-only variant. The 27B scores 87.7% zero-shot on MedQA (89.8% best-of-5), ahead of much larger general models, and the 4B fine-tunes well on small clinical datasets.
Honest caveats: the weights ship under Google's custom Gemma license, not an OSI-open license, and the headline numbers are Google-reported — independent clinical validation is still thin. It is a research and development tool, not a certified medical device.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- MedQA (4-option): 87.7% 0-shot, 89.8% best-of-5 (Google-reported)
- MedMCQA: 74.2% (Google-reported)
- MIMIC-CXR: 81% of generated reports rated likely to match or beat human decisions (Google-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Google (DeepMind Health) release, announced at Google I/O 2025; 1.5 revision shipped Jan 2026, so it is actively maintained; Strong clinical-AI community uptake: Ollama/HF variants and medical fine-tuning guides widespread; Weights open under Google's custom Gemma license, not OSI-open; training data not disclosed; MedQA numbers are Google-reported; the 4B variant's real value is fine-tuning on small datasets. License: Google Gemma license (custom, non-OSI) — commercial use permitted with restrictions, not Apache/MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Google (DeepMind Health) release, announced at Google I/O 2025; 1.5 revision shipped Jan 2026, so it is actively maintained
- · Strong clinical-AI community uptake: Ollama/HF variants and medical fine-tuning guides widespread
- · Weights open under Google's custom Gemma license, not OSI-open; training data not disclosed
- · MedQA numbers are Google-reported; the 4B variant's real value is fine-tuning on small datasets