InternVL3 38B is a visually-grounded generalist that handles both text and images with notable depth for its open-weight tier. It processes diagrams, charts, and natural images alongside text, making it capable of multi-modal reasoning without requiring proprietary access. At 38B parameters it sits in a mid-to-large range, balancing capability with the resource demands of self-hosting.