Art-Qwen3-30B-A3B-Thinking

This repository contains the Chain-of-Thought (CoT) efficient version of the Qwen3-30B-A3B-Thinking-2507 model, presented in the paper The Art of Efficient Reasoning: Data, Reward, and Optimization.

The model was trained on the taki555/DeepScaleR-Easy dataset using Reinforcement Learning (RL) strategies to incentivize accurate yet concise reasoning trajectories, addressing the computational overhead often associated with scaled CoT.

Resources

Paper: The Art of Efficient Reasoning: Data, Reward, and Optimization
Project Page: https://wutaiqiang.github.io/project/Art

Citation

@inproceedings{wu2026art,
  title={The Art of Efficient Reasoning: Data, Reward, and Optimization},
  author={Taiqiang Wu and Zenan Xu and Bo Zhou and Ngai Wong},
  year={2026},
  url={https://arxiv.org/pdf/2602.20945}
}