Med42-v2 has now been released, the latest evolution in our generative AI model built specifically for the global healthcare community. With over 70 billion parameters, Med42-v2 is designed to respond to medical prompts and questions ethically, safely, and efficiently.
Med42-v2 offers healthcare professionals and patients the ability to easily decipher medical reports, analyze test results, and streamline the healthcare experience. As an open-access suite of clinical large language models (LLMs), Med42-v2 expands access to medical knowledge and provides high-quality answers to medical questions, leveraging Meta’s Llama3 8 and 70 billion parameter models.
Med42-v2 70B surpasses GPT-4o in multiple medical benchmarks, but we recognize the growing importance of real-world performance over standardized tests. To ensure top-tier output quality, Med42-v2 70B achieves a zero-shot performance score of 80.4 on the MedQA (USMLE) benchmark, setting a new state-of-the-art among open medical LLMs. This performance, evaluated through responses to 4,000 carefully curated, publicly accessible healthcare-related questions, places Med42-v2 70B at the top of our Clinical Elo Rating Leaderboard.
For more details on the model training: Click here
Licensed on Llama 3 Community License Agreement. Read the technical report on Med42-v2 on arXiv.
Available to the healthcare community around the world, using Med42 is as simple as asking a question. With the right prompts, Med42 provides high-quality answers to short and long-form medical questions.
Med42 has been trained using M42’s industry-leading curated dataset of medical knowledge. The data comes from peer-reviewed, medically-approved sources of information including medical textbooks and literature along with datasets from national and international health agencies. As a result, Med42 achieved an impressive 72% score on the United States Medical Licensing Examination (USMLE) sample questions.
Our open-access approach enables collaboration for further development and improvement. Licensed on terms similar to Meta’s Llama 2 model, Med42 is free for non-commercial use and research – with appropriate controls on use, given the obvious risks that could arise when deploying AI in a healthcare context.
Med42 is open access and built in partnership with trusted partners including Core42 and Cerebras, to provide an unprecedented level of transparency and collaboration.
We have worked diligently with the global healthcare community to ensure the safe and ethical deployment of the Med42 technology. Our open-access approach enhances the performance of the AI model, cultivates a collaborative environment that fuels innovation among various institutions. ultimately making Med42 more open, accessible, customizable, and cost-effective.
In future, Med42 could help healthcare professionals put their energy where it matters most: on the patient.
As Med42-based technologies become regulated for widespread use, they are poised to transform the future of AI across the healthcare sector and create a direct positive impact on patient care outcomes. These could range from helping doctors diagnose rare diseases to supporting scientists in predicting the risk of patients developing certain complications.
Med42 embodies M42’s commitment to pushing the boundaries of healthcare innovation. It charts a path toward personalized, patient-centric care and accelerated global dissemination of healthcare knowledge.
The model has been developed to support the development of systems that would exponentially enhance clinical decision-making and increase access to synthesized medical knowledge for healthcare professionals, patients, and medical researchers through to regulators.
