Vision-aligned Latent Reasoning
for Multi-modal Large Language Model

 

Byungwoo Jeon¹   ·   Yoonwoo Jeong²   ·   Hyunseok Lee¹ ·   Minsu Cho^{2 3}   ·   Jinwoo Shin^{1 3   
1} KAIST   ²POSTECH   ³RLWRLD  

[project page] [arXiv]

This repository is the official implementation of Vision-aligned Latent Reasoning for Multi-modal Large Language Model (VaLR).

TODO

• Release inference code
• Release training code and data
• Release evaluation code

BibTeX

@article{jeon2026vision,
  title={Vision-aligned Latent Reasoning for Multi-modal Large Language Model},
  author={Jeon, Byungwoo and Jeong, Yoonwoo and Lee, Hyunseok and Cho, Minsu and Shin, Jinwoo},
  journal={arXiv preprint arXiv:2602.04476},
  year={2026}
}