Google is updating its open MedGemma medical AI models, releasing version 1.5 with better imaging support and a new speech-to-text tool called MedASR. The tech giant is also launching a $100,000 hackathon on Kaggle to spur development. These tools are part of Google's Health AI Developer Foundations program, designed as starting points for developers building healthcare applications.

The updated MedGemma 1.5 4B model now handles complex 3D medical data like CT scans and MRIs, plus whole-slide histopathology imaging. Google reports improved accuracy over its predecessor for classifying disease findings. A new speech recognition model, MedASR, is fine-tuned for medical dictation and pairs with MedGemma for advanced reasoning tasks, expanding the toolkit for clinical workflows.

Developers can download these models from Hugging Face or build on them using Google Cloud's Vertex AI. The new MedGemma Impact Challenge offers $100,000 in prizes for creative healthcare applications. Google emphasizes these are developer tools requiring validation and modification for specific clinical use cases, not out-of-the-box diagnostic solutions.