Google's Med-Gemini AI Scores 91.1% on USMLE, Outperforms GPT-4 by 44.5%

May 7, 2024, 02:16 AM

Google has introduced Med-Gemini, a new family of AI models developed by Google DeepMind, designed specifically for the medical field. These models, which build on the foundations of Gemini 1.0 and 1.5 LLMs, have demonstrated superior performance in healthcare AI, outperforming GPT-4 in 10 out of 14 tasks and achieving 44.5% better performance in certain benchmarks. Med-Gemini has been particularly noted for its ability to handle complex tasks within electronic health records and for achieving a 91.1% accuracy on the US Medical Licensing Exam (USMLE). The models include Med-Gemini-L 1.0, Med-Gemini-M 1.0, Med-Gemini-M 1.5, and have capabilities for 2D and 3D medical imaging, as well as genomic risk assessments, showcasing their potential to revolutionize medical diagnostics and clinical reasoning.