Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts

Nanbeige LLM Lab, Boss Zhipin · 2026

Read Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts by Nanbeige LLM Lab, Boss Zhipin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-13367.txt.

Description

Small models are often treated as narrow specialists, but Nanbeige4.1-3B aims to be a generalist at just 3 billion parameters, covering agentic behavior, code generation, and general reasoning in one model. Training blends point-wise and pair-wise reward modeling for aligned responses. Complexity-aware rewards push code that is both correct and efficient, and turn-level supervision supports deep search with long tool-use chains.

The model reliably executed up to 600 tool-call turns on complex problems. Tests show it beating prior models of similar size, including its predecessor and Qwen3-4B, and even topping the much larger Qwen3-30B-A3B on some measures. The authors present it as the first open-source small model with this breadth. Chen Yang leads the work with fourteen co-authors.

Catalog record

CREATORNanbeige LLM Lab, Boss Zhipin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-13367

Browse the full library · About the Tower of Babel