MedGemma 27B IT
Google · 2025-07-09 · 27.4B parameters
MedGemma 27B IT is Google's open-weight, instruction-tuned multimodal medical model built on Gemma 3 27B with a medically tuned SigLIP image encoder, released on July 9, 2025 as part of Health AI Developer Foundations. Beyond the 4B multimodal model's medical-image capabilities, it adds electronic-health-record (FHIR) understanding and chest X-ray anatomy localization training, and accepts text and images with at least a 128K-token context.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 130.8 |
| MedMCQA | 70.2 |
| MedQA | 85.3 |
| MedXpertQA Text | 23.7 |
| PubMedQA | 77.2 |