MedGemma 1.5 4B
2026-01-13 · Publisher : Google · Domain : Artificial Intelligence
Google Research has released MedGemma 1.5 4B, an open healthcare model that now handles 3D imaging (CT, MRI), histopathology and chest X-ray follow-up, together with the MedASR dictation model.
What changes
- Support for high-dimensional medical imaging: CT scans, MRI and whole-slide histopathology images.
- Longitudinal analysis of chest X-rays and localization of anatomical features.
- Extraction of structured data from laboratory reports.
- Improved accuracy on text, medical records and 2D images compared with MedGemma 1 4B.
- Simultaneous launch of MedASR (open medical speech recognition) and a Kaggle competition with a $100,000 prize.
### Specifications - Size: 4 billion parameters, compact enough to run offline. - Modalities: text and medical images (2D, CT/MRI slices, histopathology patches). - Availability: open weights on Hugging Face, deployment on Vertex AI. Hugging Face identifier: google/medgemma-1.5-4b-it. - Knowledge cutoff: not disclosed in the blog post. - The MedGemma 1 27B model (text) remains recommended for complex text use cases. ### Pricing Open weights model, free for research and commercial use according to the blog post. No API pricing. ### Benchmarks Scores published by the vendor, internal benchmarks: absolute accuracy of 61% versus 58% for MedGemma 1 on the classification of disease-related findings in CT (a 3-point gain). The other results are in the model card, not included here. ### Key takeaways An open entry point for prototyping medical imaging use cases on infrastructure you control. The model is a starting point to adapt, not a clinically validated device.
Links
Tags : Open Source, Google, Gemma, MedGemma, Santé