And it surpasses GPT-4 in every aspect.

Med-Gemini is a family of Gemini models fine-tuned for solving medical tasks.

Med-Gemini has set a new record score of 91.1% on the MedQA-USMLE benchmark.

The MedQA-USMLE benchmark is a dataset of multiple-choice questions in the style of the USMLE, used to evaluate the medical knowledge and reasoning capabilities of AI systems.

It also outperformed GPT-4 models by an average of 44.5% across 7 multimodal metrics.

Med-Gemini’s results are preferred over human experts 

when tackling tasks involving long texts, including:

• Summarizing medical topics
• Generating referrals to physicians
• Simplifying medical information

An AI that surpasses doctors — that’s amazing!