[Subscribe to be notified](https://services.google.com/fb/forms/hai-def-news) [Let us showcase your applications](https://services.google.com/fb/forms/hai-def-showcase/) [Give us feedback](https://services.google.com/fb/forms/hai-def-feedback/)

<br />

[### April 6th, 2026
Enhancing Medical Image Discovery: Similarity Search in the Imaging Data Commons Archive
Transform Imaging Data Commons pathology images into a searchable, discoverable archive using embedding vectors.Read More](https://developers.google.com/health-ai-developer-foundations/blogs/idc-image-search) [### Mar 26, 2026
Announcing the winners of the MedGemma Impact Challenge
The MedGemma Impact Challenge on Kaggle featured over 850 teams using Google's open medical models to solve global health challenges. Winning solutions included EpiCast for disease outbreak detection in West Africa, Sunny for skin health tracking, and FieldScreen AI for tuberculosis screening.Read More](https://blog.google/innovation-and-ai/technology/health/med-gemma-impact-challenge/) [### Jan 13, 2026
Launching MedGemma 4B 1.5 and MedGemma Impact Challenge on Kaggle
Guided by the feedback from the community, we introduced MedGemma 1.5 4B expanding support for high-dimensional imaging (CT, MRI, and whole slide histopathology), longitudinal analysis of chest X-rays, anatomical localization, and improved baseline performance in existing medical imaging and text functionality. We also launched the MedGemma Impact Challenge on Kaggle.Read More](https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr) [### Dec 19, 2025
Introducing MedASR
MedASR is a foundational speech-to-text model pre-trained on 5,000 hours of clinical audio. It can be fine-tuned for specialized contexts and paired with models like MedGemma to automate clinical documentation workflows. Read More](https://developers.google.com/health-ai-developer-foundations/medasr) [### Dec 8, 2025
Integrating MedGemma into clinical workflows just got easier!
MedGemma's toolkit now includes DICOMweb integration and an agentic FHIR navigation demo to help you integrate it into clinical workflows. Read More](https://cloud.google.com/blog/topics/developers-practitioners/integrating-medgemma-into-clinical-workflows-just-got-easier) [### July 30, 2025
Introducing LangExtract: A Gemini Powered Information Extraction Library
LangExtract is an open-source Python library that uses large language models (LLMs) to reliably and programmatically extract structured information from unstructured text. It's designed to solve the challenges of manually processing large volumes of documents like clinical notes or legal reports. We showcase Radiology use cases using Gemini. Read More](https://developers.googleblog.com/en/introducing-langextract-a-gemini-powered-information-extraction-library/) [### July 09, 2025
Expanding MedGemma Collection with 27B Multimodal and MedSigLIP
In May of this year, we expanded the HAI-DEF collection with MedGemma, a collection of generative models based on Gemma 3 that are designed to accelerate healthcare and lifesciences AI development. Today, we're proud to announce two new models in this collection. Read More](https://research.google/blog/medgemma-our-most-capable-open-models-for-health-ai-development/) [### July 05, 2025
Concluding Paris Hackathon
We invited doctors, developers and researchers to come togeher in Google Paris and prototype new medical solutions using our open models. Read More](https://blog.google/technology/health/google-france-ai-healthcare-hackathon/) [### May 29, 2025
MedGemma at Out Of Pocket (OOP) Hackathon
The weekend before its official debut at Google I/O, we provided a select group of developers in San Francisco with early access to MedGemma, challenging them to solve real-world problems in healthcare. Read More](https://www.outofpocket.health/p/oops-2025-healthcare-ai-hackathon-projects) [### May 22, 2025
Introducing MedGemma
At Google I/O, we announced MedGemma, Google's most capable open model for multimodal medical text and image comprehension. It is a collection of Gemma 3 variants (4B multimodal and 27B text-only), fine-tuned for medical tasks. Read More](https://research.google/blog/google-research-at-google-io-2025/) [### March 25, 2025
Introducing TxGemma
TxGemma is Google's first open LLM for predicting properties of therapeutics. It is a collection of Gemma 2 variants (2B, 9B, and 27B), fine-tuned for therapeutic tasks using data from the Therapeutic Data Commons. It includes TxGemma-Predict for direct task outputs and TxGemma-Chat for conversational applications. Read More](https://developers.googleblog.com/en/introducing-txgemma-open-models-improving-therapeutics-development/) [### March 17, 2025
Expanding HAI-DEF with HeAR
Our Health Acoustic Representations (HeAR) model which we published last August is now openly available as an open model in the HAI-DEF collection. Read More](https://developers.google.com/health-ai-developer-foundations/hear) [### March 14-16, 2025
HAI-DEF at MIT GrandHack 2025
We challenged researchers, developers, clinicians, and entrepreneurs to use HAI-DEF models in solving real world problems. Read More](https://hackingmedicine.mit.edu/events/grandhack-2025) [### Nov 25, 2024
Introducing Health AI Developer Foundations (HAI-DEF)
We are open-sourcing our foundation models under flexible terms of use, enabling developers and researchers to use our pretrained models at no cost, customize them, and use for product research \& development. The models are available in Google Model Garden and Hugging Face. While they are not exclusive to Google Cloud, you'll find them production-ready when deployed using Google Model Garden. Read More](https://research.google/blog/helping-everyone-build-ai-for-healthcare-applications-with-open-foundation-models/) [### Oct 21, 2024
Experimental API for 3D Computed Tomography Foundation Model
We've created Computed Tomography (CT) Foundation, a new AI tool that takes 3D CT scans and encodes them into small, information-rich numerical embeddings. This breakthrough allows researchers to build powerful new models for CT studies with significantly less data and computational resources than ever before. We've made this models available to the researchers as experimental APIs. Read More](https://research.google/blog/taking-medical-imaging-embeddings-3d/) [### Aug 19, 2024
Exploring Disease Detection Based on Bioacoustics
We've developed Health Acoustic Representations (HeAR), an AI model that learns to extract health insights from bioacoustic sounds like coughs. We show that using this embedding model new models for a wide range of diseases can be developed with less data and computational resources. We've made the model available to researchers as an experimental API. Read More](https://blog.google/technology/health/ai-model-cough-disease-detection/) [### Mar 8, 2024
Experimental APIs for Dermatology and Pathology Foundation Models
We've developed new deep learning models in Dermatology and Pathology that can encode medical images into compressed numerical vectors called embeddings. These embeddings capture an image's key features, allowing new models for various medical tasks to be developed with far less data and compute than traditional methods. We've made these models available to researchers as experimental APIs. Read More](https://research.google/blog/health-specific-embedding-tools-for-dermatology-and-pathology/) [### May 15, 2024
Exploring Medical AI with Med-Gemini
In this research we explore Med-Gemini, a family of models based on Gemini that have been fine-tuned for medical tasks. Read More](https://research.google/blog/advancing-medical-ai-with-med-gemini/) [### Aug 3, 2023
Exploring Multimodal Medical AI
In this research we explore integrating various sources like images, lab results, and patient notes to build a more comprehensive and assistive medical AI. Read More](https://research.google/blog/multimodal-medical-ai/) [### July 19, 2022
Simplified transfer learning for CXR model development
We've developed a new machine learning method to create pre-trained neural networks that convert chest X-Ray (CXR) images into data-rich vectors (embeddings). We show that with this technique model researchers can train powerful new models using far less data and computational power than traditional methods. Read More](https://research.google/blog/simplified-transfer-learning-for-chest-radiography-model-development/)