A compact vision-language model that handles both text and image inputs, producing text outputs. As a quantized (w4a16) variant, it trades some precision for reduced memory footprint and faster inference, making it more accessible on consumer hardware. It follows the Qwen3 family's instruction-following design, responding to prompts that combine visual and textual context.