Cohere has launched Aya Vision, a new multimodal AI model designed to understand text and images in 23 languages. Available on Hugging Face, Aya Vision comes in two versions: 8B and 32B parameters. The model is noted for its efficiency, outperforming larger competitors in vision-language tasks while using fewer resources. It features dynamic resizing and Pixel Shuffle technology to enhance image processing. Aya Vision can perform tasks such as image captioning, answering image-based questions, and multilingual translation. The release aims to make advancements in vision-language models accessible to the research community, and just two days post-launch, it is already trending on Hugging Face. Cohere collaborated with Kaggle to provide open weights for the model, further emphasizing its commitment to research accessibility.