Shorten After You’re Right: Lazy Length Penalties for Reasoning RL (2026.findings-acl)
Copied to clipboard
Danlong Yuan, Tian Xie, Shaohan Huang, Huishuai Zhang, Zhuocheng Gong, Chong Luo, Furu Wei, Dongyan Zhao
| Challenge: | Existing shortening methods for long reasoning models rely on additional supervision or multi-stage post-training. |
| Approach: | They propose a lazy length penalty that imposes length pressure on models without extra training stages. |
| Outcome: | The proposed method significantly reduces response length without extra training stages while maintaining or improving performance. |
Similar Papers
Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing reinforcement learning methods for large reasoning models suffer from excessive verbosity, known as "overthinking." Existing models penalize generated tokens to promote conciseness, but these methods encounter two challenges: they may develop hacking behavior in later stages of training by discarding reasoning steps. |
| Approach: | They propose a framework that steers large reasoning models toward more efficient reasoning . they prioritize correctness while imposing penalties for redundant steps . |
| Outcome: | The proposed framework reduces token usage by 69.7% on AIME24. |
Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models (2026.acl-long)
Copied to clipboard
Wei Wu, Liyi Chen, Congxi Xiao, Tianfu Wang, Qimeng Wang, Chengqiang Lu, Yan Gao, null Yiwu, Yao Hu, Hui Xiong
| Challenge: | Existing efficient reasoning methods rely on explicit length penalties for excessive verbosity on simple queries. |
| Approach: | They propose a training-time intervention that selectively suppresses redundant tokens . they find length shift occurs when models generate unnecessary reasoning on trivial inputs - a phenomenon that is often unexplored . |
| Outcome: | The proposed method reduces inference token usage by 78% while increasing accuracy compared to the initial policy and surpasses state-of-the-art efficient reasoning methods. |
LEASH: Adaptive Length Penalty and Reward Shaping for Efficient Large Reasoning Model (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to long reasoning traces are hard to tune and fail to adapt to evolving LLMs. |
| Approach: | They propose a reinforcement learning framework that optimizes the length of reasoning traces by a Lagrangian primal–dual method. |
| Outcome: | The proposed framework reduces the average reasoning length by 60% across diverse tasks while maintaining competitive performance. |
Not All Tokens Matter: Towards Efficient LLM Reasoning via Token Significance in Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) often produce unnecessarily long explanations that reduce efficiency. |
| Approach: | They propose a length-aware reward that selectively penalizes insignificance tokens . they also propose 'dynamic length control' that encourages more detailed reasoning . |
| Outcome: | The proposed method reduces response length while maintaining correctness, the authors show . it selectively penalizes insignificance tokens while maintaining accuracy . |
Optimizing Length Compression in Large Reasoning Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large Reasoning Models suffer from producing unnecessary and verbose reasoning chains. |
| Approach: | They propose a post-training method that uses a Length Reward and a Compress Reward to remove the invalid portion of the thinking process. |
| Outcome: | The proposed method reduces sequence length by 50% with only a marginal (2%) drop in accuracy. |
ReCUT: Balancing Reasoning Length and Accuracy in LLMs via Stepwise Trails and Preference Optimization (2025.findings-emnlp)
Copied to clipboard
Zhensheng Jin, Xinze Li, Yifan Ji, Chunyi Peng, Zhenghao Liu, Qi Shi, Yukun Yan, Shuo Wang, Furong Peng, Ge Yu
| Challenge: | Existing methods to train LLMs suffer from overthinking, leading to lengthy reasoning traces . Existing approaches to train large language models suffer from this problem . |
| Approach: | They propose a method to combine multiple reasoning chains for training LLMs . they use stepwise exploration and long-short switched sampling to evaluate reasoning paths . |
| Outcome: | The proposed method reduces reasoning lengths by approximately 30-50% . it also maintains or improves reasoning accuracy compared to baselines . |
Correct, Concise and Complete: Multi-stage Training For Adaptive Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) increase test-time computation, often in the form of chain-of-thought (CoT) however, reasoning traces can become unnecessarily long, increasing computation costs without improving accuracy and sometimes even degrading performance. |
| Approach: | They propose a multi-stage efficient reasoning method that combines supervised fine-tuning with reinforcement learning using an adaptive length penalty. |
| Outcome: | The proposed method reduces response length by an average of 28% for 8B models and 40% for 32B models while incurring only minor performance drops of 1.6 and 2.5 points, respectively. |
AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Models (2026.findings-acl)
Copied to clipboard
Feng Luo, Yu-Neng Chuang, Guanchu Wang, Hoang Anh Duy Le, Shaochen Zhong, Hongyi Liu, Jiayi Yuan, Yang Sui, Vladimir Braverman, Vipin Chaudhary, Xia Hu
| Challenge: | Existing approaches to distilling large language models (LLMs) are inefficient and generate excessively long chain-of-thought reasoning even for inputs that admit concise solutions. |
| Approach: | They propose a distillation framework that empowers non-reasoning LLMs to think only when necessary. |
| Outcome: | The proposed framework reduces reasoning length up to 71% with minimal accuracy loss while preserving accuracy. |
Stabilizing Efficient Reasoning with Step-Level Advantage Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models generate long and verbose reasoning traces at inference time . short context post-training alone induces substantial reasoning compression . |
| Approach: | They propose a step-level advantage selection approach that reduces reasoning length by over 30% . they propose to use GRPO without any length-aware objective to train models in a shorter context window . |
| Outcome: | The proposed approach reduces average reasoning length by over 30% while improving Pass@1 accuracy by 3.79 points over the strongest length-aware baseline. |
Step Potential Advantage Estimation: Harnessing Intermediate Confidence and Correctness for Efficient Mathematical Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to RLVR provide sparse supervision since reward arrives only after the full generation is complete. |
| Approach: | They propose a step-level reward system that extracts confidence and correctness and combines them into a Step Potential signal that explicitly estimates reasoning state at each step. |
| Outcome: | The proposed method outperforms existing methods on multiple benchmarks and improves accuracy while reducing response length. |