@misc{indiciae5f4be733c78b,
  title = {Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling},
  author = {Hongjian Zou and Yue Ge and Qi Ding and Yixuan Liao and Xiaoxin Chen},
  year = {2026},
  url = {https://arxiv.org/abs/2604.13054},
  note = {Source identifier: 2604.13054}
}