Multimodal large language models (MLLMs) have substantially advanced chart understanding and reasoning, yet their performance remains highly uneven across ch…
机构:中科院
来源:arXiv 2610.28282 | AI4Papers 论文推荐平台