A compact vision-language model that punches at a reasonable weight for its size class. It handles both text and image inputs, making it capable of tasks like image understanding and visual question answering. Being open-weight under MIT license means it's fully inspectable and deployable without usage restrictions.