MedGemma
2025-05-20 · Publisher : Google · Domain : Artificial Intelligence
Google releases MedGemma, a collection of open models derived from Gemma 3 for healthcare: 4B multimodal and 27B text at launch, complemented on July 9 by a 27B multimodal and the MedSigLIP encoder.
What changes
- MedGemma 4B multimodal and MedGemma 27B text, released on May 20, 2025, specialized in medical text and imaging.
- MedGemma 27B multimodal (interpretation of imaging and longitudinal patient records) and MedSigLIP (image-text encoder with 400 million parameters) on July 9, 2025.
- Use for report generation, questions on images and classification or retrieval of medical images.
### Specifications - Context: at least 128,000 tokens (model card). - Models usable on a single GPU; open weights (Hugging Face, Vertex Model Garden). - Use governed by the Health AI Developer Foundations terms. ### Pricing Open weights, with no usage pricing of their own. ### Benchmarks | Model | Benchmark | Score | |---|---|---| | MedGemma 4B | MedQA | 64.4% | | MedGemma 27B text | MedQA | 87.7% | | MedGemma 4B | Chest X-ray reports judged sufficiently accurate by a radiologist | 81% | Scores published by the vendor (research post of July 9). ### Key takeaways An open base for healthcare applications that require keeping data on-premises. Results must be validated before any medical use.
Links
Tags : Open Source, LLM, Google, Gemma, Santé