Papers with reasoning model
Context Reasoner: Incentivizing Reasoning Capability for Contextualized Privacy and Safety Compliance via Reinforcement Learning (2025.emnlp-main)
Copied to clipboard
Wenbin Hu, Haoran Li, Huihao Jing, Qi Hu, Ziqian Zeng, Sirui Han, Xu Heli, Tianshu Chu, Peizhao Hu, Yangqiu Song
| Challenge: | Current mitigation strategies fail to preserve contextual reasoning capabilities in risky scenarios, leading to systemic risks for legal compliance. |
| Approach: | They propose to use reinforcement learning with a rule-based reward to incentivize contextual reasoning capabilities while enhancing compliance with safety and privacy norms. |
| Outcome: | The proposed model outperforms Qwen2.5-7B-Instruct model in safety and privacy benchmarks and achieves +8.58% accuracy improvement. |
Revealing the Barriers of Language Agents in Planning (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing studies show language agents lack human-level planning abilities . limitations and mechanisms to address them remain insufficiently understood . |
| Approach: | They apply a feature attribution study to identify key factors hindering agent planning . they identify the limited role of constraints and diminishing influence of questions . |
| Outcome: | The proposed model achieves 15.6% on a real-world planning benchmark. |
Don’t Judge a Book by its Cover: Testing LLMs’ Robustness Under Logical Obfuscation (2026.eacl-long)
Copied to clipboard
| Challenge: | obfuscated questions pose significant challenges for large language models . current models parse questions without deep understanding, MIT researchers say . |
| Approach: | They propose a structure-preserving framework for logical obfuscation to test models . they use a logically equivalent framework to obliviate questions to logical equivalents . |
| Outcome: | The proposed framework is a first-of-its-kind diagnostic benchmark with 1,108 questions . obfuscation severely degrades zero-shot performance, the authors show . |
ComfyUI-R1: Exploring Reasoning Models for Workflow Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | ComfyUI-R1 is the first large reasoning model for automated workflow generation. |
| Approach: | They propose a large reasoning model for automated workflow generation that builds on curated knowledge bases and a two-stage framework to fine-tune models for cold start and reinforcement learning for incentivizing reasoning capability. |
| Outcome: | The proposed model achieves 97% format validity rate, high pass rate, node-level and graph-level F1 scores, surpassing prior state-of-the-art methods that employ leading closed-source models such as GPT-4o and Claude series. |
AdaptThink: Reasoning Models Can Learn When to Think (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large reasoning models have demonstrated remarkable capabilities in tackling complex tasks. |
| Approach: | They propose an algorithm to teach reasoning models to choose the optimal thinking mode based on problem difficulty. |
| Outcome: | The proposed algorithm reduces the average response length and improves accuracy on three math datasets. |
Counterfactual Off-Policy Training for Neural Dialogue Generation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for open-domain dialogue generation suffer from data insufficiency . a potential response inferred in hindsight is called a counterfactual reasoning . |
| Approach: | They propose to explore potential responses by counterfactual reasoning . given an observed response, the model automatically infers the outcome of an alternative policy that could have been taken . |
| Outcome: | The proposed model outperforms the HRED model and conventional learning frameworks on the DailyDialog dataset. |
KodCode: A Diverse, Challenging, and Verifiable Synthetic Dataset for Coding (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing code-focused resources typically fail to ensure either the breadth of coverage or verifiable correctness. |
| Approach: | They propose a synthetic dataset that provides high-quality, verifiable training data for Large Language Models for coding. |
| Outcome: | The proposed dataset surpasses Qwen2.5-Coder-32B-Instruct and DeepSeek-R1-Distill-Llama-70B in performance on coding benchmarks. |
Multi-step Problem Solving Through a Verifier: An Empirical Analysis on Model-induced Process Supervision (2024.findings-emnlp)
Copied to clipboard
| Challenge: | a method for process supervision has shown significant improvements in multi-step problem solving . despite the advances in process supervision, there are still easily observable mistakes in state-of-the-art LLMs. |
| Approach: | They propose a method for automating data curation by using a trained verifier to evaluate intermediate steps generated by a reasoner. |
| Outcome: | The proposed method improves the performance of PaLM 2 on math and coding tasks. |
RobustLR: A Diagnostic Benchmark for Evaluating Logical Robustness of Deductive Reasoners (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that perform deductive reasoning on inputs containing rules and statements in the English natural language do not perform consistently on the RobustLR test set. |
| Approach: | They propose a diagnostic benchmark that evaluates the robustness of language models to minimal logical edits in inputs and different logical equivalence conditions. |
| Outcome: | The proposed models do not perform consistently on the RobustLR test set. |
Internal states before wait modulate reasoning patterns (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Prior work has shown that a significant driver of performance in reasoning models is their ability to reason and self-correct. |
| Approach: | They address the question whether model’s latents preceding wait tokens contain relevant information for modulating the subsequent reasoning process. |
| Outcome: | The proposed model's latents preceding wait tokens contain relevant information for modulating the subsequent reasoning process. |
LinkQA: Synthesizing Diverse QA from Multiple Seeds Strongly Linked by Knowledge Points (2026.acl-long)
Copied to clipboard
| Challenge: | Existing training data is limited in high-quality training data, limiting the ability to produce high-performance LLMs. |
| Approach: | They propose a KP-graph-based synthesis framework that extracts KPs from QA seed data and constructs a graph of KP data from multiple seeds strongly linked by KP. |
| Outcome: | The proposed framework enables flexible control over discipline and difficulty distributions while balancing KP coverage and popularity. |
Probe Then Retrieve and Reason: Distilling Probing and Reasoning Capabilities into Smaller Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Recent research efforts have focused on distilling Large Language Models into Small Language Model (SLMs) however, the results of CoT distillation are inadequate for knowledge-intensive reasoning tasks. |
| Approach: | They propose a retrieval-based framework which distills question probing and reasoning capabilities from Large Language Models into SLMs. |
| Outcome: | The proposed framework improves probing and reasoning capabilities of large language models in knowledge-intensive reasoning tasks. |
Think Before you Write: QA-Guided Reasoning for Character Descriptions in Books (2026.findings-acl)
Copied to clipboard
| Challenge: | Character description generation is an important capability for narrative-focused applications . however, generating accurate character descriptions from long-form narratives is challenging . enabling built-in reasoning mode of current LLMs often degrades performance . |
| Approach: | They propose a framework that decouples reasoning from generation by generating a structured QA reasoning trace and a generation model that conditions on this trace to produce the final character description. |
| Outcome: | The proposed framework improves faithfulness, informativeness, and grounding over long-context baselines. |
Enhancing Multilingual Reasoning via Steerable Model Merging (2026.findings-acl)
Copied to clipboard
Zhuoran Li, Rui Xu, Jian Yang, Junnan Liu, Zhijun Chen, Qianren Mao, Hongcheng Guo, Jiaheng Liu, Likang Xiao, Ming LI, Xiaojie Wang
| Challenge: | Model merging is an effective technique for composing the capabilities of a multilingual model and a reasoning model. |
| Approach: | They propose a model merging framework that modulates the contribution of each source model. |
| Outcome: | Experiments show that the proposed model merging framework outperforms strong baselines on multilingual reasoning benchmarks across 21 different languages. |
Decoupled Reasoning with Implicit Fact Tokens (DRIFT): A Dual-Model Framework for Efficient Long-Context Inference (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing solutions to integrate extensive, dynamic knowledge into Large Language Models (LLMs) are constrained by finite context windows, retriever noise, or the risk of catastrophic forgetting. |
| Approach: | They propose a dual-model architecture that explicitly decouples knowledge extraction from the reasoning process by compressing document chunks into implicit fact tokens conditioned on the query. |
| Outcome: | The proposed architecture significantly outperforms strong baselines among comparably sized models on long-context tasks while maintaining inference accuracy. |
Embodied-Reasoner: Synergizing Visual Search, Reasoning, and Action for Embodied Interactive Tasks (2026.acl-long)
Copied to clipboard
Wenqi Zhang, Mengna Wang, Gangao Liu, Huixin Xu, Yiwei Jiang, Yongliang Shen, Guiyang Hou, Zhe Zheng, Hang Zhang, Xin Li, Jiajun Liu, Weiming Lu, Peng Li, Yueting Zhuang
| Challenge: | Recent advances in reasoning models have demonstrated remarkable capabilities on mathematical and coding tasks, but their effectiveness in embodied domains remains largely unexplored. |
| Approach: | They propose a reasoning model for interactive embodied tasks that synthesizes 9.3k coherent Observation-Thought-Action trajectories containing 64k ego-centric images and 90k diverse reasoning processes. |
| Outcome: | The proposed model outperforms existing visual reasoning models by +9%, 24%, and +13% on long-horizon tasks. |