A multimodal reasoning model that takes text as input and produces text output, built on a combination of CLIP and FLAN-T5 XL architectures. It inherits FLAN-T5 XL's instruction-following strengths while incorporating CLIP's vision-language alignment capabilities, though its text-only I/O interface suggests it operates on pre-encoded or abstracted representations. Details about its specific task focus and fine-tuning objectives are limited.