AHA · LOOPED TRANSFORMERSubmit a paper ↗
← Back to the collection
Full-stack recurrence ·

Nanbeige4.2-3B

Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model

Nanbeige Lab, Chen Yang, Chengrui Huang, Fufeng Lan, Hanhui Chen, Hao Zhou, Huatong Song, Jiaqi Cao, Jiaying Zhu, Jinlin Niu, Kai Wang, Lisheng Huang, Qiliang Liang, Ran Le, Ruixiang Feng, Shuang Sun, Tao Gu, Tao Zhang, Tianyu Luo, Yang Song, Yun Xing, Yuntao Wen, Ziyao Xu, Zongchao Chen, Zongqiang Li

A released language model that applies recurrent computation across a shared decoder stack.

Inside the method

[ 22-layer decoder stack ] × 2

Simplified conceptual schematic. Consult the paper for the complete architecture.

Recurrence family
Full-stack recurrence
Depth control
Two passes in released config
KV / state strategy
See model implementation

Reading note

The published configuration specifies 22 layers and 2 loops. Other modes supported by the code do not imply that the released weights use them.

Sources checked 2026-09-15. This catalog does not imply independent reproduction.

Cite this work

@misc{nanbeige2026,
  title = {Nanbeige4.2-3B: Unlocking Agentic Capabilities in a Compact Model},
  author = {Nanbeige Lab and Chen Yang and Chengrui Huang and Fufeng Lan and Hanhui Chen and Hao Zhou and Huatong Song and Jiaqi Cao and Jiaying Zhu and Jinlin Niu and Kai Wang and Lisheng Huang and Qiliang Liang and Ran Le and Ruixiang Feng and Shuang Sun and Tao Gu and Tao Zhang and Tianyu Luo and Yang Song and Yun Xing and Yuntao Wen and Ziyao Xu and Zongchao Chen and Zongqiang Li},
  year = {2026},
  eprint = {2607.22083},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.22083}
}

Suggest a correction ↗