arXiv AI

When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

When2Think introduces a post‑training framework that dynamically allocates reasoning depth in Large Reasoning Models based on instance difficulty. The method uses Instance‑level Difficulty‑Aware Control (IDAC) to shape rewards with pre‑computed accuracy and token usage statistics, enabling stable, critic‑free optimization without learned reward models. Experiments on mathematical benchmarks show that When2Think improves accuracy‑efficiency trade‑offs, achieving higher Pass@3 scores while reducing token usage compared to baseline models.

arXiv AI
Aug 21

Learning When to Think: Adaptive Reasoning for Test-Time Compute Allocation

arXiv:2608. 20256v1 Announce Type: new Abstract: Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones.

By Gijs Kassenaar, Zhao Yang, Vincent Fran\c{c}ois-Lavet
arXiv AI
Sep 28

Nice Fold or Hero Call: Learning Budget-Efficient Thinking under Policy-Dependent Solvability

The paper introduces Budget‑Efficient Thinking (BET), a two‑stage framework that treats adaptive reasoning as a computational investment, aligning solve‑or‑fold decisions with expected return rather than perceived difficulty. BET learns three distinct behaviors: concise short solves for easy queries, early abstention (nice fold) when further reasoning is unlikely to pay off, and allocating sufficient compute (hero call) for hard‑but‑solvable questions. Experiments on seven benchmarks with three base models show BET cuts reasoning tokens by 54% while boosting accuracy by up to 3.2%, and it transfers effectively to scientific QA and logical reasoning tasks.

By Zhaomeng Zhou, Lan Zhang, Junyang Wang, Mu Yuan, Songlin Liu, Tingzhao Li, Yiqing Hu, Yumeng Zhao
arXiv AI
Sep 25

To Think or Not to Think: Allocating Reasoning Where It Helps

The paper introduces CARE, a contrastive accuracy reward estimation method that adaptively adjusts reasoning length for large language models. By comparing beneficial length adjustments from online sampled responses, CARE applies adaptive length rewards within Group Relative Policy Optimization without extra hyperparameters or inference cost. Experiments on multiple reasoning benchmarks show that CARE improves Pass@1 by up to 4% while reducing reasoning length by 37%, achieving higher token efficiency.

By Zhengdong He, Yunfan Zhou, Jianguo Yao, Haibing Guan, Xijun Li
arXiv AI
Oct 15

Efficiently Learning To Reason or Not to Reason: Root-token Policy Optimization for Adaptive Thinking

The paper introduces Root-token Policy Optimization (RPO), a framework that lets large reasoning models decide when to engage in explicit reasoning by training only the initial root token. RPO uses group-relative rewards and advantages to focus on this branching point, dramatically cutting training overhead and VRAM usage. Experiments across various model families show that RPO achieves difficulty-aware adaptive thinking with roughly 2% of the compute required by previous adaptive-reasoning methods.

By Taehyeon Kim, Hyunsoo Lee, Youngsoo Jang, Moontae Lee
arXiv AI
Jun 8

DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling

arXiv:2606. 07108v1 Announce Type: new Abstract: Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking".

By Tengyao Tu, Yulin Li, Hui-Ling Zhen, Libo Qin, Zhoujun Wei, Jinghua Piao, Zhuotao Tian, Yong Li, Min Zhang
arXiv Machine Learning
Jun 9

The Easy, the Hard, and the Learnable: Confidence and Difficulty-Adaptive Policy Optimization for LLM Reasoning

arXiv:2606. 07950v1 Announce Type: new Abstract: RL with verifiable rewards can substantially improve LLM reasoning, yet standard GRPO-style training often treats easy, hard, and learnable questions alike through uniform sampling and weighting, leading to inefficient compute allocation.

By Zhanke Zhou, Xiangyu Lu, Chentao Cao, Brando Miranda, Tongliang Liu, Bo Han, Sanmi Koyejo
arXiv AI
Oct 28

Difficulty-Estimated Policy Optimization

Difficulty-Estimated Policy Optimization (DEPO) is a new framework that improves the efficiency and robustness of reasoning alignment in Large Reasoning Models. It introduces an online Difficulty Estimator that filters training data before rollouts, ensuring computational resources focus on high‑learning‑potential samples. Empirical results show DEPO achieves about 1.86× faster rollouts than DAPO and outperforms GRPO on benchmarks at similar training costs.

By Yu Zhao, Fan Jiang, Tianle Liu, Bo Zeng, Xintong Wang, Jiaxuan Teng, Yu Liu, Yichao Du