Google Launches Open AI Models MedGemma 1.5 and MedASR for Healthcare Research
Google has launched MedGemma 1.5 and MedASR, two new AI models designed for healthcare research. Unlike some competitors that keep healthcare AI behind paid enterprise services, Google released both models publicly for researchers and developers. As a result, the wider community can experiment, improve workflows, and build medical AI tools more freely.
MedGemma 1.5 targets medical images and text
MedGemma 1.5 is the latest version of Google’s medical vision-language model built to analyse medical images alongside written information. In addition, it can interpret scans, answer questions about visual medical data, and support research-focused tasks.
According to Google Research, the updated version brings improved multimodal reasoning and stronger performance on complex medical imagery. Moreover, it offers flexibility, so researchers can fine-tune it for specialised datasets and specific study goals.
The model supports multiple medical imaging formats, including radiology scans and other clinically relevant visuals. For example, Google said MedGemma 1.5 can help with image-based question answering, report drafting, and structured data extraction. The company stressed that it is not meant to provide diagnoses or treatment advice and should only be used as a support tool in research and development settings.
MedASR focuses on clinical speech recognition and access
Alongside MedGemma 1.5, Google introduced MedASR, an automatic speech recognition model designed specifically for healthcare environments. Similarly, MedASR focuses on accurate transcription of clinical conversations, including medical terms, varied accents, and real-world background noise.
Google said the model aims to reduce transcription errors that often occur when general-purpose speech recognition systems are used in medical contexts. Therefore, teams can use it to transcribe doctor-patient discussions, generate clinical notes, and convert dictated reports into text. Additionally, the company said MedASR can adapt to different healthcare settings through fine-tuning for specific workflows and documentation standards.
Google said all versions of MedGemma and MedASR are available through Hugging Face and the Vertex AI platform. Meanwhile, developers can access documentation and tutorials via the MedGemma GitHub repository.
