A compact vision-language model that handles both text and image inputs, producing text outputs. At 12B parameters, it sits in a middle ground — capable enough for multimodal reasoning tasks without the resource demands of larger models. The BF16 format suggests it's tuned for efficient inference on modern hardware.