We present NextFlow, a unified decoder-only autoregressive transformer trained on 6 trillion interleaved text-image discrete tokens. By leveraging a unified vision representation within a unified autoregressive architecture, NextFlow natively activates multimodal understanding and generation capabilities, unlocking abilities of image editing, interleaved content and video generation. Motivated by the distinct nature of modalities - where text is strictly sequential and images are inherently hierarchical - we retain next-token prediction for text but adopt next-scale prediction for visual generation. This departs from traditional raster-scan methods, enabling the generation of 1024x1024 images in just 5 seconds - orders of magnitude faster than comparable AR models. We address the instabilities of multi-scale generation through a robust training recipe. Furthermore, we introduce a prefix-tuning strategy for reinforcement learning. Experiments demonstrate that NextFlow achieves state-of-the-art performance among unified models and rivals specialized diffusion baselines in visual quality.
@article{arxiv.2601.02204,
title = {NextFlow: Unified Sequential Modeling Activates Multimodal Understanding and Generation},
author = {Huichao Zhang and Liao Qu and Yiheng Liu and Hang Chen and Yangyang Song and Yongsheng Dong and Shikun Sun and Xian Li and Xu Wang and Yi Jiang and Hu Ye and Bo Chen and Yiming Gao and Peng Liu and Akide Liu and Zhipeng Yang and Qili Deng and Linjie Xing and Jiyang Liu and Zhao Wang and Yang Zhou and Mingcong Liu and Yi Zhang and Qian He and Xiwei Hu and Zhongqi Qi and Jie Shao and Zhiye Fu and Shuai Wang and Fangmin Chen and Xuezhi Chai and Zhihua Wu and Yitong Wang and Zehuan Yuan and Daniel K. Du and Xinglong Wu},
journal= {arXiv preprint arXiv:2601.02204},
year = {2026}
}