Google Unveils Med-Gemini Family of Multimodal Medical AI Models Claimed to Outperform GPT-4

On Tuesday, Google unveiled its new family of artificial intelligence (AI) models focused on the medical field. Called Med-Gemini, these AI models are not available for human use, but the tech giant has published a pre-print version of its research paper that highlights its capabilities and methodologies. The company claims that AI models outperform GPT-4 models in benchmark tests. One of the notable features of this particular AI model is its long context capabilities, which allow it to process and analyze health records and scientific articles.

The survey paper is currently in preprint and has been published on arXiv, an open access online repository for scientific articles. Jeff Dean, Principal Scientist, Google DeepMind and Google Research, said in a post at X (formerly known as Twitter), “I am very excited about the possibilities of these models to help clinicians provide better care, as well as help patients better understand their medical conditions. In my opinion, AI for healthcare will be one of the most impactful application areas for AI.”

Med-Gemini AI models are built on Gemini 1.0 and Gemini 1.5 LLM. There are four models in total — Med-Gemini-S 1.0, Med-Gemini-M 1.0, Med-Gemini-L 1.0 and Med-Gemini-M 1.5. All models are multimodal and can provide text, image and video output. The models are also integrated with web search, which the company claims is enhanced through self-learning to make the models “more factually accurate, reliable and nuanced” when displaying results for complex clinical reasoning tasks.

Additionally, the AI ​​model has been fine-tuned for improved performance during long context processing, the company claims. Higher-quality processing of long context would mean that the chatbot could provide more accurate and precise answers, even when the questions are not asked perfectly or when it has to process a long document of medical records.

According to data shared by Google, AI Med-Gemini models outperformed OpenAI’s GPT-4 models on the GeneTuring dataset on text reasoning tasks. The Med-Gemini-L 1.0 also scored 91.1 percent MedQA (USMLE) accuracy, even beating its own older Med-PaLM 2 model by 4.5 percent. It should be noted that the AI ​​model is not publicly available or in beta testing. The company is likely to further improve the model before making it public.