ModelsUS vs China12:12Chinese contentSciFigQual-Bench:面向科学图像质量评估的全文本基准评估论文图表质量?SciFigQual-Bench这个新基准让GPT-5.6-Sol驱动的SFQ-Agent打分,误差比直问模型低43%,更靠谱。#SciFigQual-Bench#GPT-5.6-Sol#SFQ-Agent#图像质量评估aarXiv cs.AI@Zihan Deng et al. (6)SourceRead laterReadWorth followingUsefulFollow SciFigQual-Bench