Google introduced its new family of artificial intelligence (AI) models focused on the medical domain on Tuesday. Dubbed Med-Gemini, these AI models are not available for people to use, but the tech giant has published a pre-print version of its research paper which highlights its capabilities and methodologies. The company claims that the AI models surpass GPT-4 models in benchmark testing. One of the notable features of this particular AI model is its long-context abilities that allow it to process and analyse health records and research papers.

The research paper is currently in the pre-print stage and is published on arXiv, an open-access online repository of scholarly papers. Jeff Dean, Chief Scientist, Google DeepMind and Google Research, said in a post on X (formerly known as Twitter), “I’m very excited about the possibilities of these models to help clinicians deliver better care, as well as to help patients better understand their medical conditions. AI for healthcare is going to be one of the most impactful application domains for AI, in my opinion.”

Med-Gemini AI models are built on top of Gemini 1.0 and Gemini 1.5 LLM. There are a total of four models — Med-Gemini-S 1.0, Med-Gemini-M 1.0, Med-Gemini-L 1.0, and Med-Gemini-M 1.5. All of the models are multimodal and can provide text, image, and video outputs. The models are also integrated with web search, which the company claims has been improved through self-training to make the models “more factually accurate, reliable, and nuanced” when showing results for complex clinical reasoning tasks.

Further, the AI model is fine-tuned for improved performance during long-context processing, claims the company. A higher quality long-context processing would mean the chatbot can provide more accurate and pinpointed answers even when the questions are not perfectly queried or when it has to process a long document of medical records.

As per data shared by Google, Med-Gemini AI models have outperformed OpenAI’s GPT-4 models in the GeneTuring dataset on text-based reasoning tasks. Med-Gemini-L 1.0 has also scored 91.1 percent accuracy on MedQA (USMLE), even outperforming its own older model Med-PaLM 2 by 4.5 percent. Notably, the AI model is not available in public or in beta testing. The company likely will improve the model further before bringing it into the public domain.