LARGE REASONING MODELS

How Reliable are Confidence Estimators for Large Reasoning Models? A Systematic Benchmark on High-Stakes Domains (2026.eacl-long)

Challenge:	Existing approaches to Chain-of-Thought reasoning are often degraded because they disregard the model’s intrinsic reasoning dependency.
Approach:	They propose a self-guided pruning framework that leverages the model’s intrinsic likelihood landscape to identify segments that are extraneous to its specific reasoning pattern.
Outcome:	The proposed framework reduces output length while maintaining or improving accuracy on multiple benchmarks.

Copied to clipboard

Reza Khanmohammadi, Erfan Miahi, Simerjot Kaur, Charese Smiley, Ivan Brugere, Kundan S Thind, Mohammad M. Ghassemi

Beyond Token Length: Step Pruner for Efficient and Accurate Reasoning in Large Language Models (2026.findings-acl)

Challenge:	Large Reasoning Models often struggle with confidence calibration, authors say . authors: accurate confidence scores are essential to build trustworthy systems .
Approach:	They propose a Reasoning Model Confidence estimation benchmark to assess LRM confidence . the benchmark is constructed from 347,496 reasoning traces from six popular LRMs .
Outcome:	The proposed benchmark compares ten different representation-based methods on a wide range of architectures.

Copied to clipboard

Canhui Wu, Qiong Cao, Chang Li, Zhenfang Wang, Chao Xue, Yuwei Fan, Wei Xi, Xiaodong He

Think Outside the Policy: In-Context Steered Policy Optimization (2026.findings-acl)

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.

Copied to clipboard

Hsiu-Yuan Huang, Chenming Tang, Weijie Liu, Clive Bai, Saiyong Yang, Yunfang Wu

CiPO: Counterfactual Unlearning for Large Reasoning Models through Iterative Preference Optimization (2026.acl-long)

Challenge:	Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods exhibit limited exploration due to reliance on on-policy rollouts which are limited to the current policy’s distribution, resulting in narrow trajectory diversity.
Approach:	They propose a framework that leverages the in-context learning capability of Large Reasoning Models to provide expert guidance using existing datasets.
Outcome:	The proposed framework improves RLVR performance and training stability on mathematical reasoning benchmarks.

Copied to clipboard

Junyi Li, Yongqiang Chen, Ningning Ding

Optimizing Length Compression in Large Reasoning Models (2026.acl-long)

Challenge:	Existing methods to unlearning large reasoning models do not remove unwanted knowledge from CoT traces or interfere with the reasoning process.
Approach:	They propose a framework that targets the CoT reasoning in Large Reasoning Models by generating a valid counterfactual reasoning trace for preference tuning.
Outcome:	Experiments on large LRMs show that CiPO completely removes knowledge from the intermediate CoT steps and the final answer while preserving the reasoning abilities of LRM.

Copied to clipboard

Zhengxiang Cheng, Dongping Chen, Mingyang Fu, Tianyi Zhou

CAT: Confidence-Adaptive Thinking for Efficient Reasoning of Large Reasoning Models (2026.acl-industry)

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.

Copied to clipboard

Qizhi Jiang, Shuo Wang, Pei Ke, Yuhang Song, Ke Qin

Safety in Large Reasoning Models: A Survey (2025.findings-emnlp)

Challenge:	Existing compression methods for large reasoning models rely on uniform length reduction or coarse-grained difficulty estimation, often leading to performance degradation on difficult problems.
Approach:	They propose a framework that incorporates model’s intrinsic self-certainty signals as confidence into the preference optimization process, which autonomously modulates reasoning lengths based on problem difficulty.
Outcome:	The proposed framework outperforms state-of-the-art models on reasoning accuracy across multiple benchmarks on different base models.

Copied to clipboard

Cheng Wang, Yue Liu, Baolong Bi, Duzhen Zhang, Zhong-Zhi Li, Yingwei Ma, Yufei He, Shengju Yu, Xinfeng Li, Junfeng Fang, Jiaheng Zhang, Bryan Hooi

ThinkPilot: Steering Reasoning Models via Automated Think-prefixes Optimization (2026.findings-eacl)

Challenge:	Large Reasoning Models (LRMs) have a high level of advanced reasoning capabilities, but they are vulnerable and vulnerable.
Approach:	This paper presents the first comprehensive survey of Large Reasoning Models . it explores the new safety risks, attacks, and defense strategies specific to LRMs based on reasoning .
Outcome:	The proposed study examines the safety and security risks of large reasoning models.

Copied to clipboard

Sunzhu Li, Zhiyu Lin, Jiale Zhao, Shuling Yang, Chen Wei

Reasoning’s Razor: Reasoning Improves Accuracy but Hurts Recall at Critical Operating Points in Safety and Hallucination Detection (2026.eacl-long)

Challenge:	Large Reasoning Models (LRMs) are powerful but still suffer from inefficient and off-target reasoning.
Approach:	They propose a training-free framework that automatically optimizes Large Reasoning Models' reasoning by generating think-prefixes that evolve driven by a taxonomy of reasoning behaviors.
Outcome:	The proposed framework significantly improves accuracy-length trade-off for efficient reasoning, drastically improves safety and improves instruction following.

Copied to clipboard

Atoosa Chegini, Hamid Kazemi, Garrett Souza, Maria Safi, Yang Song, Samy Bengio, Sinead Williamson, Mehrdad Farajtabar

RiT: Rubrics-in-Thinking Reinforcement Learning for Improved Reasoning in Large Language Models (2026.findings-acl)

Challenge:	a new study examines the suitability of reasoning for precision-sensitive classification tasks . false positives carry severe operational consequences, such as blocking legitimate queries .
Approach:	They propose to use reasoning for classification tasks under low false positive rate regimes . they find that Think On improves overall accuracy, but performs poorly at low FPRs a .
Outcome:	The proposed reasoning-augmented generation model outperforms self-verbalized confidence in precision-sensitive deployments.

Copied to clipboard

Xiaobin Tian, Shuai Yuan, Muyun Ding, Haonan Chen, Xiaoxi Jiang

DRP: Distilled Reasoning Pruning with Mathematical Skill-aware Step Decomposition for Efficient Large Reasoning Models (2026.findings-acl)

Challenge:	Large Reasoning Models benefit from generating intermediate reasoning steps alongside final answers.
Approach:	They propose a framework to introduce thinking-rubric supervision into intermediate reasoning.
Outcome:	The proposed framework outperforms outcome-only RL baselines on reasoning-intensive and open-ended tasks.

Copied to clipboard

Yuxuan Jiang, Dawei Li, Francis Ferraro

Green & Sustainable NLP

Challenge:	Existing solutions to this problem are inference-time pruning and tuning-based distillation.
Approach:	They propose a framework that combines inference-time pruning with tuning-based distillation to enable efficient and accurate reasoning.
Outcome:	The proposed framework reduces token usage while improving accuracy on GSM8K and AIME tokens while avoiding performance drop.

From Reasoning to Answer: Empirical, Attention-Based and Mechanistic Insights into Distilled DeepSeek R1 Models (2025.emnlp-main)

Copied to clipboard

Jue Zhang, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang

R-TOFU: Unlearning in Large Reasoning Models (2025.emnlp-main)

Challenge:	Large Reasoning Models generate explicit reasoning traces alongside final answers . the extent to which these traces influence answer generation remains unclear .
Approach:	They conduct empirical evaluation of Large Reasoning Models that include explicit reasoning . they also show that answer tokens attend substantially to reasoning tokens .
Outcome:	The results show that including explicit reasoning improves answer quality across domains . they also show that answer tokens attend substantially to reasoning tokens - the authors .

Copied to clipboard

Sangyeon Yoon, Wonje Jeung, Albert No

From "Aha Moments" to Controllable Thinking: Toward Meta-Cognitive Reasoning in LRMs via Decoupled Reasoning and Control (2026.acl-long)

Challenge:	Large Reasoning Models embed private or copyrighted information in their final answers and throughout multi-step chain-of-thought (CoT) traces.
Approach:	They propose a benchmark for Large Reasoning Models that augments existing unlearning tasks with realistic CoT annotations and step-wise metrics that expose residual knowledge invisible to answer-level checks.
Outcome:	The proposed benchmark shows that answer-only objectives leave substantial forget traces in reasoning.

Copied to clipboard

Rui Ha, Rui Pu, Chaozhuo Li, Li Sun, Sen Su

TrigReason: Trigger-Based Collaboration between Small and Large Reasoning Models (2026.findings-acl)

Challenge:	Large Reasoning Models exhibit step-by-step reasoning, reflection, and backtracking, but these behaviors are often unregulated, leading to overthinking.
Approach:	They propose a meta-cognitive reasoning framework that decouples reasoning from control to enable independent optimization of control strategies.
Outcome:	Experiments show that the proposed model improves efficiency and accuracy across reasoning benchmarks.

Copied to clipboard

Yi Zhao, Yajuan Peng, Cam-Tu Nguyen, Zuchao Li, Xiaoliang Wang, Xiaoming Fu, Hai Zhao

Three Minds, One Legend: Jailbreak Large Reasoning Model with Adaptive Stacked Ciphers (2026.findings-acl)

Challenge:	Large Reasoning Models suffer from high inference latency due to autoregressive reasoning . SpecReason adopts a polling-based design that repeatedly invokes the LRM for verification at every step .
Approach:	They propose a trigger-based collaborative reasoning framework that delegates most reasoning to the SRM and activates LRM intervention only when necessary.
Outcome:	The proposed framework reduces latency and API cost by 73.3% under edge–cloud conditions.

Copied to clipboard

Nguyen Viet Anh, Shiqian Zhao, Gia Dao, Runyi Hu, Yi Xie, Xiaobao Wu, Anh Tuan Luu

Teach Small Models to Reason by Curriculum Distillation (2025.emnlp-main)

Challenge:	Existing jailbreak methods struggle to balance effectiveness with robustness against adaptive safety mechanisms.
Approach:	They propose a novel approach that targets Large Reasoning Models through an adaptive encryption pipeline designed to overwhelm their reasoning capabilities.
Outcome:	The proposed approach achieves an attack success rate of 85.6% on OpenAI GPT-o4-mini, outperforming state-of-the-art baselines by a significant margin of 17.2%.

Copied to clipboard

Wangyi Jiang, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

OpenHuEval: Evaluating Large Language Model on Hungarian Specifics (2025.findings-acl)

Challenge:	Large Reasoning Models (LRMs) show strong System-2-style reasoning, but at the cost of significant computational overhead.
Approach:	They propose a two-stage curriculum distillation framework which builds a robust internal problem-solving student model and then teaches the student model to externalize this knowledge as explicit reasoning.
Outcome:	The proposed model outperforms single-stage baselines on mathematical benchmarks and significantly outperformed LRMs on complex tasks.

Copied to clipboard

Haote Yang, Xingjian Wei, Jiang Wu, Noémi Ligeti-Nagy, Jiaxing Sun, Yinfan Wang, Győző Zijian Yang, Junyuan Gao, Jingchao Wang, Bowen Jiang, Shasha Wang, Nanjun Yu, Zihao Zhang, Shixin Hong, Hongwei Liu, Wei Li, Songyang Zhang, Dahua Lin, Lijun Wu, Gábor Prószéky, Conghui He

ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning (2025.emnlp-main)

Challenge:	Recent advances in Large Language Models (LLMs) represent significant strides toward artificial general intelligence (AGI).
Approach:	They introduce OpenHuEval, the first benchmark for LLMs focusing on the Hungarian language and specifics.
Outcome:	The framework reveals intrinsic patterns and mechanisms of LLMs in non-English languages, with Hungarian serving as an example.

Copied to clipboard

Ziqing Qiao, Yongheng Deng, Jiali Zeng, Dong Wang, Lai Wei, Guanbo Wang, Fandong Meng, Jie Zhou, Ju Ren, Yaoxue Zhang

A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaos (2025.findings-acl)

Challenge:	Existing methods for fine-tuning-based compression suffer from verbose outputs, increasing computational overhead.
Approach:	They propose a framework to generate concise reasoning chains using Confidence Injection and Early Stopping.
Outcome:	The proposed framework reduces the length of the model by up to 50% while maintaining high task accuracy.

Copied to clipboard

Yang Yao, Xuan Tong, Ruofan Wang, Yixu Wang, Lujundong Li, Liang Liu, Yan Teng, Yingchun Wang

PAM: Enhancing General Alignment of Large Reasoning Models through Priority-Aware Metacognition (2026.acl-long)

Challenge:	Large Reasoning Models (LRMs) have advanced beyond traditional Large Language Models, yet they pose heightened safety risks.
Approach:	They propose a first jailbreak attack targeting Large Reasoning Models . they exploit a Chaos Machine component to transform attack prompts with diverse one-to-one mappings based on the reasoning chain .
Outcome:	The proposed attack exploits the unique vulnerabilities of LRMs by integrating a Chaos Machine. success rates of the mousetrap attack are as high as 96%, 86% and 98% respectively.

Copied to clipboard

Zhihao Xu, Fuzhen Yang, Liang Lin, Xiting Wang

Large Language and Reasoning Models are Shallow Disjunctive Reasoners (2025.acl-long)

Challenge:	Existing studies indicate that System-2 thinking alone does not transfer to the general alignment domain.
Approach:	They propose to use priority-aware metacognition to help LRMs understand human preferences and monitor and regulate their thinking process.
Outcome:	The proposed model improves general alignment performance by 10 points on helpfulness and harmless benchmarks.

Copied to clipboard

Irtaza Khalid, Amir Masoud Nourollah, Steven Schockaert

Conflicts Make Large Reasoning Models Vulnerable to Attacks (2026.findings-acl)

Challenge:	Large Language Models (LLMs) struggle with systematic reasoning on out-of-distribution (OOD) tasks.
Approach:	They propose to use a set of constraints to measure OOD generalization to create large reasoning models that can be leveraged to solve real-world problems.
Outcome:	The proposed models outperform their LLM counterparts in single-path reasoning tasks but struggle in multi-path setting.

Copied to clipboard

Honghao Liu, Chengjin Xu, Xuhui Jiang, Cehao Yang, Shengming Yin, Zhengwu Ma, Lionel Ni, Jian Guo

Stress-Testing the Reasoning Competence of Language Models With Formal Proofs (2025.findings-emnlp)

Challenge:	Large Reasoning Models have demonstrated outstanding capabilities in solving complex reasoning tasks by incorporating step-by-step chain-of-thought (CoT) reasoning.
Approach:	They evaluate three large reasoning models that perform explicit and coherent reasoning under conflicting objectives and use them to evaluate their performance.
Outcome:	The proposed models perform explicit and coherent reasoning before producing their outputs, improving problem-solving and multi-step decision making.

Copied to clipboard

Konstantine Arkoudas, Serafim Batzoglou

START: Self-taught Reasoner with Tools (2025.emnlp-main)

Challenge:	a new battery of challenging but tractable logical inference tasks is developed on ProofGrid . the tasks include proof writing and proof checking across propositional and equational logic .
Approach:	They propose a new battery of challenging but tractable logical inference tasks on ProofGrid . they introduce two new tasks: proof inpainting and proof gap-filling .
Outcome:	The proposed model performs well on top-tier models but also shows systematic failure modes.

Copied to clipboard

Chengpeng Li, Mingfeng Xue, Zhenru Zhang, Jiaxi Yang, Beichen Zhang, Bowen Yu, Binyuan Hui, Junyang Lin, Xiang Wang, Dayiheng Liu

Logic-Thinker: Teaching Large Language Models to Think more Logically. (2025.findings-emnlp)

Challenge:	Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in complex reasoning through long chain-of-thought, yet they struggle with precise computations and algorithmic operations.
Approach:	They propose a training-free approach that activates LRMs’ latent tool-use capabilities through artificial hints and a framework that enables models to learn effective tool utilization through diverse hint patterns and rejection-based data synthesis.
Outcome:	Experiments show that START significantly improves state-of-the-art LRMs across challenging benchmarks, including competition-level mathematics (AMC23: 95.0%, AIME24: 75.6%) and graduate-level science questions (GPQA: 64.6%).

Copied to clipboard

Chengyao Wen, Qiang Cheng, Shaofei Wang, Zhizhen Liu, Deng Zhao, Lei Liang

Understanding LLM Reasoning for Abstractive Summarization (2026.findings-acl)

Challenge:	Recent Large Reasoning Models (LRMs) have demonstrated the ability to generate long chains of thought (LongCoT) LongCoT still faces challenges such as redundancy and logical incoherence.
Approach:	They propose a neural-symbolic reasoning framework that generates chains of thought . they propose Logic-Thinker, which transforms symbolic solvers into chains of thoughts .
Outcome:	The proposed framework outperforms models fine-tuned with ThinkerCoT on logic reasoning tasks.

Copied to clipboard

Haohan Yuan, Haopeng Zhang

Text Generation

Challenge:	Explicit reasoning strategies improve reference-based quality, but weaken factual grounding, whereas implicit reasoning in LRMs shows the opposite tendency.
Approach:	They adapt general reasoning strategies to the summarization setting and conduct a large-scale comparative study of 8 reasoning strategies and 3 Large Reasoning Models (LRMs) they find a trade-off between summary quality and factual faithfulness.
Outcome:	The proposed reasoning strategies and 3 Large Reasoning Models (LRMs) are compared with 8 reasoning strategies across 8 datasets.

S*: Test Time Scaling for Code Generation (2025.findings-emnlp)

Copied to clipboard

Dacheng Li, Shiyi Cao, Chengkun Cao, Xiuyu Li, Shangyin Tan, Kurt Keutzer, Jiarong Xing, Joseph E. Gonzalez, Ion Stoica

Programming Languages in NLP

Challenge:	S* is the first hybrid test-time scaling framework that significantly improves the coverage and selection accuracy of generated code.
Approach:	They propose a hybrid test-time scaling framework that augments parallel scaling with sequential scaling to further increase the performance.
Outcome:	The proposed framework outperforms existing scaling approaches in large-scale modeling and reasoning models.

GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts (2026.findings-acl)

Copied to clipboard

Wenhao Zeng, Xuteng Zhang, Yuling Shi, Chao Hu, Yuting Chen, Beijun Shen, Xiaodong Gu

Towards a Mechanistic Understanding of Large Reasoning Models: A Survey of Training, Inference, and Failures (2026.acl-long)

Challenge:	Existing routing strategies rely on local token probabilities or post-hoc verification, introducing significant inference overhead.
Approach:	They propose a step-wise collaboration framework that generates only the first token of each reasoning step and routes it to a larger model only when initial token entropy exceeds a threshold.
Outcome:	The proposed approach reduces inference latency while preserving accuracy.

Copied to clipboard

Yi Hu, Jiaqi Gu, Ruxin Wang, Zijun Yao, Hao Peng, Xiaobao Wu, Jianhui Chen, Muhan Zhang, Liangming Pan

Mapping the Minds of LLMs: A Graph-Based Analysis of Reasoning LLMs (2025.emnlp-main)

Challenge:	Recent research has shown that reinforcement learning can elicit intriguing emergent reasoning behaviors.
Approach:	They propose a comprehensive survey of the mechanistic understanding of large reasoning models . they organize findings into three core dimensions: 1) training dynamics, 2) reasoning mechanisms, and 3) unintended behaviors.
Outcome:	This paper synthesizes the mechanistic understanding of large reasoning models into three dimensions . authors outline a roadmap for future studies including improved interpretability and methodologies .

Copied to clipboard

Zhen Xiong, Yujun Cai, Zhecheng Li, Yiwei Wang

Understanding the Thinking Process of Reasoning Models: A Perspective from Schoenfeld’s Episode Theory (2025.emnlp-main)

Challenge:	Large Reasoning Models (LRMs) often display unstable behaviors, e.g., hallucinating unsupported premises, overthinking simple tasks, and displaying higher sensitivity to prompt variations.
Approach:	They propose a graph-based analytical framework that clusters long, verbose CoT outputs into semantically coherent reasoning steps, then constructs directed reasoning graphs to capture contextual and logical dependencies among these steps.
Outcome:	The proposed framework enables quantitative evaluation of internal reasoning structure and quality beyond conventional metrics and provides practical insights for prompt engineering and cognitive analysis of LLMs.

Copied to clipboard

Ming Li, Nan Zhang, Chenrui Fan, Hong Jiao, Yanbin Fu, Sydney Peters, Qingshu Xu, Robert Lissitz, Tianyi Zhou

Linguistics & Cognitive NLP

Challenge:	Large Reasoning Models (LRMs) generate extensive chain-of-thought reasoning, but we lack a principled framework for understanding how these thoughts are structured.
Approach:	They propose a method to analyze the reasoning traces of Large Reasoning Models using Schoenfeld’s Episode Theory.
Outcome:	The proposed framework provides a theoretically grounded methodology for interpreting LRM cognition and enables future work on more controllable and transparent reasoning systems.

How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study (2026.acl-long)

Copied to clipboard

Zhexin Zhang, Xian Qi Loye, Victor Shea-Jay Huang, Junxiao Yang, Qi Zhu, Shiyao Cui, Fei Mi, Lifeng Shang, Yingkang Wang, Hongning Wang, Minlie Huang

Think Natively: Unlocking Multilingual Reasoning with Consistency-Enhanced Reinforcement Learning (2026.acl-long)

Challenge:	Large Reasoning Models have achieved remarkable success on reasoning-intensive tasks, but their enhanced reasoning capabilities do not translate to improved safety performance.
Approach:	They propose to use supervised fine tuning to enhance the safety of Large Reasoning Models.
Outcome:	The proposed method improves the safety of large reasoning models on reasoning-intensive tasks.

Copied to clipboard

Xue Zhang, Yunlong Liang, Fandong Meng, Songming Zhang, Kaiyu Huang, Yufeng Chen, Xu Jinan, Jie Zhou

Controllable LLM Reasoning via Sparse Autoencoder-Based Steering (2026.acl-long)

Challenge:	Current Large Reasoning Models exhibit two critical limitations when processing non-English languages: (1) They struggle to maintain input-output language consistency; (2) They generally perform poorly with wrong reasoning paths and lower answer accuracy compared to English.
Approach:	They propose a language-consistency reward and a cross-lingual thinking alignment reward to improve the model's interpretability and accuracy.
Outcome:	The proposed model achieves nearly 100% language consistency and superior performance on two multilingual benchmarks (MMATH and PolyMath).

Copied to clipboard

Yi Fang, Wenjie Wang, Mingfeng Xue, Boyi Deng, Fengli Xu, Dayiheng Liu, Fuli Feng

HAUNTATTACK: When Attack Follows Reasoning as a Shadow (2026.findings-acl)

Challenge:	Existing methods struggle to control fine-grained reasoning strategies due to conceptual entanglement in LRMs’ hidden states.
Approach:	They propose to decompose strategy-entangled hidden states into a disentangled feature space by using Sparse Autoencoders to identify the few strategy-specific features from the vast pool of SAE features.
Outcome:	The proposed method outperforms existing methods by 15% in control effectiveness.

Copied to clipboard

Jingyuan Ma, Rui Li, Zheng Li, Junfeng Liu, Heming Xia, Lei Sha, Zhifang Sui

ReasonRank: Empowering Passage Ranking with Strong Reasoning Ability (2026.acl-long)

Challenge:	Emerging Large Reasoning Models (LRMs) excel in mathematical and reasoning tasks, showcasing remarkable capabilities.
Approach:	They propose a framework that embeds harmful instructions into reasoning questions . they evaluate 11 LRMs and observe an average attack success rate of over 70% .
Outcome:	The proposed framework improves reasoning models by 13 percentage points over baseline.

Copied to clipboard

Wenhan Liu, Xinyu Ma, Weiwei Sun, Yutao Zhu, Yuchen Li, Dawei Yin, Zhicheng Dou

NeuReasoner: Towards Explainable, Controllable, and Unified Reasoning via Mixture-of-Neurons (2026.acl-long)

Challenge:	Existing rerankers perform poorly in complex ranking scenarios due to the scarcity of reasoning-intensive training data.
Approach:	They propose an automated reasoning-intensive training framework which generates high-quality training labels from training queries and passages.
Outcome:	The proposed model outperforms baselines significantly and achieves much lower latency than the pointwise reranker.

Copied to clipboard

Haonan Dong, Kehan Jiang, Haoran Ye, Wenhao Zhu, Zhaolu Kang, Guojie Song

Red Teaming Large Reasoning Models (2026.acl-long)

Challenge:	Existing Large Reasoning Models (LRMs) lack explainability and controllability . Existing models target isolated levels without unification, while relying on RL .
Approach:	They propose an explainable, controllable, and unified reasoning framework driven by MoN.
Outcome:	The proposed framework achieves performance gains of 27.0% while reducing token consumption by 19.6% 63.3%.

Copied to clipboard

Jiawei Chen, Yang Yang, Chao Yu, Yu Tian, Zhi Cao, Xue Yang, Linghao Li, Hang Su, Zhaoxia Yin

ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding (2026.findings-acl)

Challenge:	Large Reasoning Models (LRMs) have emerged as a powerful advancement in multi-step reasoning tasks, but they introduce safety and reliability risks, such as CoT-hijacking and prompt-induced inefficiencies.
Approach:	They propose a unified benchmark to assess the trustworthiness of Large Reasoning Models.
Outcome:	The proposed benchmark evaluates truthfulness, safety and efficiency on 26 models.

Copied to clipboard

Sangjun Song, Minjae Oh, Seungkyu Lee, Sungmin Jo, Yohan Jo

When Models Reason in Your Language: Controlling Thinking Language Comes at the Cost of Accuracy (2025.findings-emnlp)

Challenge:	Recent advances in Large Reasoning Models (LRMs) have demonstrated remarkable capabilities across various tasks.
Approach:	They propose a system that stops reasoning when the margin between continuation token and lt;/think gt; narrows.
Outcome:	The proposed model reduces thinking token usage by 30% and improves accuracy by 8% while reducing thinking tokens by 72%.

Copied to clipboard

Jirui Qi, Shan Chen, Zidi Xiong, Raquel Fernández, Danielle Bitterman, Arianna Bisazza

When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models (2026.findings-acl)

Challenge:	Recent Large Reasoning Models (LRMs) with thinking traces have shown strong performance on English reasoning tasks.
Approach:	They evaluate two leading LRMs with thinking traces on established benchmark XReasoning and propose directions for future research.
Outcome:	The proposed models often revert to English or produce fragmented reasoning in other languages, revealing a substantial gap in the capability of thinking in non-English languages.

Copied to clipboard

Yingzhi Mao, Chunkang Zhang, Junxiang Wang, Xinyan Guan, Boxi Cao, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun

Think Less, Know More: State-Aware Reasoning Compression with Knowledge Guidance for Efficient Reasoning (2026.findings-acl)

Challenge:	Existing methods often apply coarse-grained constraints over entire reasoning trajectories . Existing approaches often apply unsafe constraints, causing unsafe outputs .
Approach:	They propose a trajectory-level training framework that mitigates Self-Jailbreak . they propose 'chain-of-guardrail' to mitigate self-jailbreak by targeting step-level interventions .
Outcome:	The proposed framework mitigates Self-Jailbreak by targeting step-level interventions while maintaining reasoning ability.

Copied to clipboard

Yi Sui, Chaozhuo Li, Dawei Song

FoE: Forest of Errors Makes the First Solution the Best in Large Reasoning Models (2026.acl-long)

Challenge:	Existing CoT compression methods struggle to balance accuracy and efficiency . long CoT reasoning also introduces an overthinking phenomenon, authors say .
Approach:	They propose a framework that performs step-wise CoT compression by modeling stage-specific redundancy sources and integrating with a retrieval-augmented guidance.
Outcome:	The proposed framework reduces average response length by 59.9% while improving accuracy by 4.8 points over existing methods.

Copied to clipboard

Kehan Jiang, Haonan Dong, Zhaolu Kang, Zhengzhou Zhu, Guojie Song

PARIF: Pushing the Pareto Frontier of Instruction Following and Reasoning with Curriculum Reinforcement Learning (2026.acl-long)

Challenge:	Recent Large Reasoning Models (LRMs) have demonstrated remarkable success in complex reasoning tasks.
Approach:	They propose a self-guided efficient reasoning framework that reduces FoE by pruning subs.
Outcome:	The proposed model outperforms eight competitive baselines while reducing token consumption by 37.7% 70.4%.

Copied to clipboard

Rongchuan Mu, Zexin Wang, Qianyu Wang, MingHua Ma, Zekun Wang, Ming Liu, Bing Qin

Marco-o1 v2: Towards Widening The Distillation Bottleneck for Reasoning Models (2025.acl-long)

Challenge:	Existing alignment methods struggle to balance general reasoning with instruction-following (IF) this is hindered by dependency on teacher models, reward hacking, and reasoning-answer inconsistencies.
Approach:	They propose a two-stage curriculum learning framework based on Reinforcement Learning from Verifiable Rewards to enhance both IF and general reasoning capabilities.
Outcome:	The proposed framework outperforms leading models on six representative IF tasks while achieving a 21.25% relative average improvement over the original model.

Copied to clipboard

Huifeng Yin, Yu Zhao, Minghao Wu, Xuanfan Ni, Bo Zeng, Huaiyu.wh Huaiyu.wh, Tianqi Shi, Liangying Shao, Chenyang Lyu, Longyue Wang, Weihua Luo, Kaifu Zhang

Graph-R1: Incentivizing the Zero-Shot Graph Learning Capability in LLMs via Explicit Reasoning (2025.emnlp-main)

Challenge:	Recent efforts to distill large reasoning models into smaller lightweight models have shown competitive performances.
Approach:	They propose to distill long Chain-of-Thought data to improve SFT and RL methods by constructing data from scratch using Monte Carlo Tree Search.
Outcome:	The proposed method significantly improves reasoning performance on various benchmarks such as math (GSM8K, MATH, AIME).

Copied to clipboard

Yicong Wu, Guangyue Lu, Yuan Zuo, Huarong Zhang, Junjie Wu

NEAT: Neuron-Based Early Exit for Large Reasoning Models (2026.findings-acl)

Challenge:	Recent advances in Large Reasoning Models (LLMs) provide a zero-shot alternative via explicit, long chain-of-thought reasoning.
Approach:	They propose a GNN-free approach that reformulates graph tasks as textual reasoning problems solved by LRMs.
Outcome:	The proposed approach outperforms state-of-the-art baselines in zero-shot settings, producing interpretable and effective predictions.

Copied to clipboard

Kang Liu, YongKang Liu, Xiaocui Yang, Peidong Wang, Wen Zhang, Shi Feng, Yifei Zhang, Daling Wang

BloomEval: A Bloom’s Cognitive Taxonomy-Based Benchmark for Evaluating LRMs via Cognitive Hierarchy Trace (2026.findings-acl)

Challenge:	Existing approaches to reduce overthinking require additional rollout computation or externally labeled datasets.
Approach:	They propose a Neuron-based Early reAsoning exiT framework that monitors neuron-level activation dynamics to enable training-free early exits.
Outcome:	The proposed framework reduces the amount of reasoning steps generated by LRMs while maintaining accuracy.

Copied to clipboard

Zhiyi Duan, Lei Gao, Jiangshan Guan, Qi Wang, Rui Liu

SafeKey: Amplifying Aha-Moment Insights for Safety Reasoning (2025.emnlp-main)

Challenge:	Existing benchmarks for Large Reasoning Models rely on answer correctness, but fail to assess the structural coherence and cognitive soundness of the reasoning process itself.
Approach:	They propose a framework that maps a model's reasoning trajectory onto hierarchical cognitive levels and an annotation pipeline to ensure a scalable yet reliable annotation pipeline.
Outcome:	The proposed framework detects hierarchy jumps, breaks, and overthinking errors and enables scalable yet reliable annotation.

Copied to clipboard

Kaiwen Zhou, Xuandong Zhao, Jayanth Srinivasa, Gaowen Liu, Aosong Feng, Dawn Song, Xin Eric Wang

REST: Stress Testing Large Reasoning Models by Asking Multiple Problems at Once (2026.acl-long)

Challenge:	Large Reasoning Models (LRMs) introduce a new paradigm of explicitly reasoning before answering, but they pose great safety risks against harmful queries and adversarial attacks.
Approach:	They propose a safety aha moment that activates safety reasoning and leads to a safe response.
Outcome:	The proposed model can generalize to unseen jailbreak prompts while maintaining general abilities.

Copied to clipboard

Zhuoshi Pan, Qizhi Pei, Yu Li, Zinan Tang, QiYao Sun, H. Vicky Zhao, Conghui He, Lijun Wu

SpecCoT: Accelerating Chain-of-Thought Reasoning through Speculative Exploration (2025.findings-emnlp)

Challenge:	Recent Large Reasoning Models (LRMs) lack a narrow evaluation paradigm . a single-question evaluation setup suffers from two major limitations .
Approach:	They propose a stress-testing framework that exposes LRMs to multiple problems simultaneously.
Outcome:	The proposed framework outperforms existing models on reasoning benchmarks and state-of-the-art models.

Copied to clipboard

Junhan Shi, Yijia Zhu, Zhenning Shi, Dan Zhao, Qing Li, Yong Jiang

ARM2: Adaptive Reasoning Model with Vision Understanding and Executable Code (2026.findings-acl)

Challenge:	Large Reasoning Models suffer from high inference latency due to lengthy reasoning chains.
Approach:	They propose a collaborative framework that combines large and small models for effective reasoning.
Outcome:	The proposed framework reduces inference latency by 1.7-4.1 while maintaining comparable accuracy to standard large model inference.

Copied to clipboard

Jian Xie, Zhendong Chu, Aoxiao Zhong, Kai Zhang, Mingzhe Han, Xing Fan, Jialie Shen, Qingsong Wen

Multimodality

Challenge:	Large Reasoning Models suffer from the "over-thinking" problem, causing performance degradation.
Approach:	They propose a unified model that balances reasoning performance and efficiency across multiple formats through a reinforcement learning framework augmented with length-aware optimization.
Outcome:	The proposed model reduces token costs while preserving performance compared to traditional models.

Think Better, Not Longer: Token-Level Marginal Utility for Efficient Reasoning in Large Reasoning Models (2026.acl-long)

Copied to clipboard

Jiawei Li, Yang Gao, Huashan Sun, Chong Feng

ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments (2026.acl-long)

Challenge:	Large reasoning models (LRMs) generate explicit Chain-of-Thought rationales, but often suffer from "overthinking".
Approach:	They propose a unified training framework to synthesize concise reasoning chains by identifying tokens that reduce the model’s likelihood of the correct answer.
Outcome:	Experiments on deepSeek-R1-Distill-Qwen backbones show that MUTO yields better efficiency-accuracy Pareto frontier.

Copied to clipboard

Yuquan Wang, Mi Zhang, Yining Wang, Geng Hong, Mi Wen, Xiaoyu You, Min Yang

Reasoning-to-Defend: Safety-Aware Reasoning Can Defend Large Language Models from Jailbreaking (2025.emnlp-main)

Challenge:	Existing defenses for Large Reasoning Models (LRMs) depend on costly fine-tuning and additional expert knowledge, which limits their scalability.
Approach:	They propose an inference-time safeguard for Large Reasoning Models that injects safety aha moments into the reasoning process to guide the model towards harmless yet helpful reasoning.
Outcome:	The proposed safeguard outperforms nine existing safeguards while avoiding common exaggerated safety issues.

Copied to clipboard

Junda Zhu, Lingyong Yan, Shuaiqiang Wang, Dawei Yin, Lei Sha

Robustness in NLP

Challenge:	Large Reasoning Models (LLMs) have demonstrated impressive performances across diverse domains, but how their safety benefits from enhanced reasoning capabilities against jailbreak queries remains unexplored.
Approach:	They propose a safety-aware reasoning paradigm that integrates a pivot token-based safety-based reasoning mechanism into LLMs’ generation process.
Outcome:	The proposed model improves the safety of large language models against jailbreak queries while minimizing attacks and maintaining the original performance.

Quantifying and Understanding Uncertainty in Large Reasoning Models (2026.acl-long)

Copied to clipboard

Yangyi Li, Chenxu Zhao, Mengdi Huai

TTVS: Boosting Self-Exploring Reinforcement Learning via Test-time Variational Synthesis (2026.findings-acl)

Challenge:	Existing methods for estimating generation uncertainty do not provide finite-sample guarantees for reasoning-answer generation.
Approach:	They propose a method that provides the uncertainty of the reasoning-answer structure with statistical guarantees.
Outcome:	The proposed method disentangles reasoning quality from answer correctness while establishing theoretical guarantees for efficient explanation methods.

Copied to clipboard

Sikai Bai, Haoxi Li, Jie Zhang, Yongjiang Liu, Song Guo

Low-Resource NLP

Challenge:	Existing test-time methods are limited in specialized or novel domains where supervision is prohibitively expensive or unavailable.
Approach:	They propose a framework that augments training stream from unlabeled test queries.
Outcome:	Extensive experiments show TTVS outperforms state-of-the-art RL-based techniques on unlabeled test-time data.

Pru-CoT: Towards Efficient Reasoning Distillation via Pruning Chain-of-Thought (2026.findings-acl)

Copied to clipboard

Han Liu, Shuotian Ma, Hui Li, Xiaotong Zhang, Fenglong Ma, Hong Yu

Word Salad Chopper: Reasoning Models Waste A Ton Of Decoding Budget On Useless Repetitions, Self-Knowingly (2025.emnlp-main)

Challenge:	Existing heuristics fail to capture global causal logic due to rigid rules and limited search spaces.
Approach:	They propose a framework that extracts the essential logical structure from reasoning chains.
Outcome:	Experiments show that Pru-CoT models generate more compact reasoning paths compared to models trained on verbose data.

Copied to clipboard

Wenya Xie, Shaochen Zhong, Hoang Anh Duy Le, Zhaozhuo Xu, Jianwen Xie, Zirui Liu

Think How to Think: Mitigating Overthinking with Autonomous Difficulty Cognition in Large Reasoning Models (2026.acl-long)

Challenge:	Large Reasoning Models (LRMs) are often bottlenecked by the high cost of output tokens.
Approach:	They propose a lightweight, turnkey component for Large Reasoning Models that is minimally invasive to its reasoning trajectory.
Outcome:	The proposed component is lightweight and low overhead, and lacks semantic value.

Copied to clipboard

Yongjiang Liu, Haoxi Li, Xiaosong Ma, Jie Zhang, Song Guo

AdaMix: Adaptive Mixing for Short and Long Reasoning Adapters (2026.acl-long)

Challenge:	Recent Large Reasoning Models (LRMs) excel at complex reasoning tasks but often suffer from overthinking.
Approach:	They propose a two-stage fine-tuning strategy that progressively inspires LRMs’ difficulty cognition and redundancy cognition of LRM.
Outcome:	The proposed model significantly reduces inference costs by over 70% on easy tasks and 40% on complex ones without compromising performance.

Copied to clipboard

Hao Luo, Xiao Yan, Xinyan Li, Qiming Zeng, Yuhao Lin, Shanshan Feng, Hao Wang, Jiawei Jiang

Green & Sustainable NLP

Challenge:	Existing methods for large reasoning models have improved efficiency but still face limitations such as conflicting objectives and limited adaptability.
Approach:	They propose an adaptive reasoning framework that applies a uniform, computation-intensive deep reasoning strategy to all problems.
Outcome:	The proposed framework reduces the average response length of DeepSeek-R1-Distill-Qwen-7B by 54.9% while improving accuracy by up to 4.8% on five mathematical datasets.

AlgBench: To What Extent Do Large Reasoning Models Understand Algorithms? (2026.findings-acl)

Copied to clipboard

Henan Sun, Kaichi Yu, Yuyao Wang, Bowen Liu, Xunkai Li, Rong-Hua Li, Nuo Chen, Jia Li

SAT: Balancing Reasoning Accuracy and Efficiency with Stepwise Adaptive Thinking (2026.acl-long)

Challenge:	Existing benchmarks for algorithmic reasoning fail to answer a critical question: do LRMs master algorithmic thinking? Empirical evaluations on leading LRM models reveal substantial performance heterogeneity, while models perform well on non-optimized tasks, accuracy drops sharply to around 49% on globally optimized algorithms.
Approach:	They propose an algorithm-centric benchmark that evaluates large reasoning models under an algorithmic paradigm.
Outcome:	Empirical evaluations on leading LRMs reveal substantial performance heterogeneity . models perform well on non-optimized tasks, accuracy drops sharply to around 49% .

Copied to clipboard

Weiyang Huang, Xuefeng Bai, Kehai Chen, Xinyang Chen, Yibin Chen, Weili Guan, Min Zhang

CoTJudger: A Graph-Driven Framework for Automatic Evaluation of Chain-of-Thought Efficiency and Redundancy in LRMs (2026.findings-acl)

Challenge:	Large Reasoning Models (LRMs) produce excessively long Chains of Thought (COT) Existing solutions that improve token efficiency but sacrifice fine-grained control can disrupt the logical integrity of the reasoning process.
Approach:	They propose a framework that performs step-level, difficulty-aware pruning while preserving the core reasoning structure.
Outcome:	Experiments show that SAT reduces reasoning tokens by 40% while maintaining or improving accuracy.

Copied to clipboard

Siyi Li, Jiajun Shi, Shiwen Ni, Ge Zhang, Shuaimin Li, Shijian Wang, Zhoufutu Wen, Yizhi LI, Hamid Alinejad-Rokny, Jiaheng Liu, Min Yang, Wenhao Huang

Green & Sustainable NLP

Challenge:	Existing evaluations emphasize final accuracy or coarse token counts, and lack automated tools to separate essential logic from structural redundancy.
Approach:	They propose a graph-driven framework that quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path needed to reach a correct solution.
Outcome:	Evaluating 21 LRMs, the proposed framework quantifies reasoning efficiency by converting free-form CoTs into directed dependency graphs and extracting the Shortest Effective Path (SEP) needed to reach a correct solution.

PolitNuggets: Benchmarking Agentic Discovery of Long-Tail Political Facts (2026.acl-long)

Copied to clipboard

Yifei Zhu

Rose-SQL: Role-State Evolution Guided Structured Reasoning for Multi-Turn Text-to-SQL (2026.acl-long)

Challenge:	Large Reasoning Models (LRMs) are embedded in agentic frameworks and are under-evaluated.
Approach:	They propose a multilingual benchmark for agentic information synthesis using PolitNuggets . they standardize evaluation with an optimized Supervisor–Searcher multi-agent system .
Outcome:	The proposed model can discover and synthesize "long-tail" facts from dispersed sources.

Copied to clipboard

Le Zhou, Feng Yao, Fengcai Qiao, Bo Xu, Fangyuan Wang, Boyan Xu

Programming Languages in NLP

Challenge:	Existing approaches to multi-turn Text-to-SQL tasks rely on unstable APIs or expensive fine-tuning.
Approach:	They propose a training-free framework that leverages small-scale LRMs through in-context learning to enable accurate context-dependent parsing.
Outcome:	The proposed framework outperforms in-context learning baselines at the 4B scale and surpasses state-of-the-art models at the 8B and 14B scales.

ReasonAny: Incorporating Reasoning Capability to Any Model via Simple and Effective Model Merging (2026.acl-long)

Copied to clipboard

Junyao Yang, Chen Qian, Wen Shen, Yong Liu, Jing Shao, Dongrui Liu