Keye VL 8B Preview is a vision-language model that punches at the intersection of visual understanding and text generation, despite its name suggesting multimodal input, it currently accepts only text. As a preview release, it carries an experimental character — capable and exploratory, but not yet a finished product. Its open-weight nature makes it inspectable and adaptable for developers willing to work with early-stage models.