A multimodal reasoning model that accepts both text and image inputs, producing text outputs. It operates at 32 billion parameters, placing it in the mid-to-large weight class. As an open-weight release from Nvidia, it can be downloaded and run locally, giving practitioners direct control over deployment.