Figure 2.

Illustration of textual and visual inputs to multimodal large language models (MLLMs). We input these questions along with their corresponding images into each of the six MLLMs and collected their responses.