InternVL-U has been officially released as open source, enabling integrated understanding, reasoning, generation, and editing with 4B parameters.

The Shanghai Artificial Intelligence Laboratory, in collaboration with several universities, has open-sourced the multimodal integrated model InternVL-U, achieving unified capabilities in understanding, reasoning, generation, and editing with only 4B parameters. The model employs an architecture of "unified contextual modeling + modal-specific modularization + decoupled visual representation," outperforming 14B-level models in complex scenarios such as text rendering and scientific reasoning. It is fully open-source and provides complete inference code and evaluation tools.