voyage-multimodal-3, launched by Voyage AI, is a multimodal embedding model that vectorizes text and images (including screenshots of PDFs, slides, and tables) while capturing key visual features. This advancement significantly enhances document retrieval accuracy for rich visual and textual information within knowledge bases, making it important for RAG and semantic search applications. On multimodal retrieval tasks, voyage-multimodal-3 achieves an average improvement of 19.63% in retrieval accuracy compared to other models.