LLMs and MLLMs show different trends in model performance when provided with different input modalities and excel at different ASCII art categories. Better training strategies or model architectures ...