AURORA-LM: Autoencoding Unified Representation for Continuous-Latent Diffusion Language Modeling

Jiajun Liang*, Yucheng Liao*, Yukang Cao*, Jiazhe Wei, Ken Li, Wende Tan, Jiankun Zhang, ZY Cui, Jingkang Yang, Liucheng Guo, Shiqi Yang, B. Yang, Caifeng Shan, Ziwei Liu, and Chenyang Si†
¹PRLab, Nanjing University   ²S-Lab, Nanyang Technological University   ³Imperial College London
* Equal contribution   † Corresponding author

📄 arXiv   |   🌐 Project Page   |   🤗 Model Weights (Coming Soon)

• [2026-08-01] The official AURORA-LM repository is available with paper overview materials.
• [2026-08-03] The paper is now available on arXiv.

AURORA-LM is a continuous-latent diffusion language model that separates the construction of a decodable text representation from the modeling of its generative distribution. It preserves a high-capacity, decodable text latent and designs the diffusion model to learn the resulting distribution directly.

• arXiv paper
• Project page

• Training and inference code
• Model checkpoints

@misc{liang2026auroralmautoencodingunifiedrepresentation,
  title={AURORA-LM: Autoencoding Unified Representation for Continuous-Latent Diffusion Language Modeling},
  author={Jiajun Liang and Yucheng Liao and Yukang Cao and Jiazhe Wei and Ken Li and Wende Tan and Jiankun Zhang and ZY Cui and Jingkang Yang and Liucheng Guo and Shiqi Yang and B. Yang and Caifeng Shan and Ziwei Liu and Chenyang Si},
  year={2026},
  eprint={2608.02602},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2608.02602}
}

For questions and collaborations, please contact:

• Jiajun Liang
• Chenyang Si