ICLR 2026 accepted papers on Large language models
1,219 of the 5,358 papers accepted at ICLR 2026 (22.8%) are on Large language models, compared with 20.0% at ICLR 2025. They fall into 7 topics.
See ICLR 2026 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Reasoning and chain of thought · 347 papers
- When More is Less: Understanding Chain-of-Thought Length in LLMs (5 citations)
- ReTool: Reinforcement Learning for Strategic Tool Use in LLMs (3 citations)
- Learning to Generate Unit Test via Adversarial Reinforcement Learning (2 citations)
- RM-R1: Reward Modeling as Reasoning (2 citations)
- AnesSuite: A Comprehensive Benchmark and Dataset Suite for Anesthesiology Reasoning in LLMs (1 citation)
- Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning (1 citation)
- Deep Think with Confidence (1 citation)
- DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoning (1 citation)
- DynaGuard: A Dynamic Guardian Model With User-Defined Policies (1 citation)
- Dynamic Early Exit in Reasoning Models (1 citation)
- GRACE: Generative Representation Learning via Contrastive Policy Optimization (1 citation)
- J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning (1 citation)
and 335 more in the interactive map.
Language, safety and interpretability · 288 papers
- Neologism Learning for Controllability and Self-Verbalization (7 citations)
- EditLens: Quantifying the Extent of AI Editing in Text (3 citations)
- SPRIG: Improving Large Language Model Performance by System Prompt Optimization (3 citations)
- Align to Misalign: Automatic LLM Jailbreak with Meta-Optimized LLM Judges (1 citation)
- AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraint (1 citation)
- Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training (1 citation)
- Don't Throw Away Your Beams: Improving Consistency-based Uncertainties in LLMs via Beam Search (1 citation)
- Doxing via the Lens: Revealing Location-related Privacy Leakage on Multi-modal Large Reasoning Models (1 citation)
- GuidedBench: Measuring and Mitigating the Evaluation Discrepancies of In-the-wild LLM Jailbreak Methods (1 citation)
- Mixture of Cognitive Reasoners: Modular Reasoning with Brain-Like Specialization (1 citation)
- On the Impossibility of Separating Intelligence from Judgment: The Computational Intractability of Filtering for AI Alignment (1 citation)
- SafeDialBench: A Fine-Grained Safety Evaluation Benchmark for Large Language Models in Multi-Turn Dialogues with Diverse Jailbreak Attacks (1 citation)
and 276 more in the interactive map.
Agents, code and math · 213 papers
- LLMs Get Lost In Multi-Turn Conversation (10 citations)
- Multi-turn Evaluation of Anthropomorphic Behaviours in Large Language Models (8 citations)
- MCP-Bench: Benchmarking Tool-Using LLM Agents with Complex Real-World Tasks via MCP Servers (4 citations)
- Can Large Language Models Match the Conclusions of Systematic Reviews? (2 citations)
- DeepResearch Bench: A Comprehensive Benchmark for Deep Research Agents (2 citations)
- SimpleToM: Exposing the Gap between Explicit ToM Inference and Implicit ToM Application in LLMs (2 citations)
- VCWorld: A Biological World Model for Virtual Cell Simulation (2 citations)
- VERINA: Benchmarking Verifiable Code Generation (2 citations)
- An Agentic Framework with LLMs for Solving Complex Vehicle Routing Problems (1 citation)
- BED-LLM: Intelligent Information Gathering with LLMs and Bayesian Experimental Design (1 citation)
- CodeSense: a Real-World Benchmark and Dataset for Code Semantic Reasoning (1 citation)
- FutureX: An Advanced Live Benchmark for LLM Agents in Future Prediction (1 citation)
and 201 more in the interactive map.
Instruction and fine-tuning · 119 papers
- Persona Features Control Emergent Misalignment (2 citations)
- ATGen: Adversarial Reinforcement Learning for Test Case Generation (1 citation)
- Detecting Data Contamination from Reinforcement Learning Post-training for Large Language Models (1 citation)
- Hey, That's My Model! Introducing Chain & Hash, An LLM Fingerprinting Technique (1 citation)
- KaLM-Embedding-V2: Superior Training Techniques and Data Inspire A Versatile Embedding Model (1 citation)
- Log Probability Tracking of LLM APIs (1 citation)
- Superficial Safety Alignment Hypothesis (1 citation)
- AFD-INSTRUCTION: A Comprehensive Antibody Instruction Dataset with Functional Annotations for LLM-Based Understanding and Design
- Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence
- AWM: Accurate Weight-Matrix Fingerprint for Large Language Models
- BA-LoRA: Bias-Alleviating Low-Rank Adaptation to Mitigate Catastrophic Inheritance in Large Language Models
- Be Careful When Fine-tuning On Open-Source LLMs: Your Fine-tuning Data Could Be Secretly Stolen!
and 107 more in the interactive map.
Retrieval and knowledge · 111 papers
- When to use Graphs in RAG: A Comprehensive Analysis for Graph Retrieval-Augmented Generation (4 citations)
- CounselBench: A Large-Scale Expert Evaluation and Adversarial Benchmarking of Large Language Models in Mental Health Question Answering (3 citations)
- DeepRAG: Thinking to Retrieve Step by Step for Large Language Models (3 citations)
- Cancer-Myth: Evaluating Large Language Models on Patient Questions with False Presuppositions (2 citations)
- WebShaper: Agentically Data Synthesizing via Information-Seeking Formalization (2 citations)
- Attributing Response to Context: A Jensen–Shannon Divergence Driven Mechanistic Study of Context Attribution in Retrieval-Augmented Generation (1 citation)
- From Single to Multi-Granularity: Toward Long-Term Memory Association and Selection of Conversational Agents (1 citation)
- Parameters vs. Context: Fine-Grained Control of Knowledge Reliance in Language Models (1 citation)
- RAEE: A Robust Retrieval-Augmented Early Exit Framework for Efficient Inference (1 citation)
- RAS: Retrieval-And-Structuring for Knowledge-Intensive LLM Generation (1 citation)
- SmartChunk Retrieval: Query-Aware Chunk Compression with Planning for Efficient Document RAG (1 citation)
- A Fano-Style Accuracy Upper Bound for LLM Single-Pass Reasoning in Multi-Hop QA
and 99 more in the interactive map.
Alignment and preferences · 93 papers
- Preference Leakage: A Contamination Problem in LLM-as-a-judge (2 citations)
- Cultivating Pluralism In Algorithmic Monoculture: The Community Alignment Dataset (1 citation)
- SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safety (1 citation)
- Search Arena: Analyzing Search-Augmented LLMs (1 citation)
- Align Once, Benefit Multilingually: Enforcing Multilingual Consistency for LLM Safety Alignment
- Alignment through Meta-Weighted Online Sampling: Bridging the Gap between Data Generation and Preference Optimization
- Antislop: A Comprehensive Framework for Identifying and Eliminating Repetitive Patterns in Language Models
- Benchmarking Overton Pluralism in LLMs
- Beyond Binary Preferences: A Principled Framework for Reward Modeling with Ordinal Feedback
- Data Selection for LLM Alignment Using Fine-Grained Preferences
- Discern Truth from Falsehood: Reducing Over-Refusal via Contrastive Refinement
- Disentangling Length Bias in Preference Learning via Response-Conditioned Modeling
and 81 more in the interactive map.
AI and society · 48 papers
- Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents (22 citations)
- GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks (6 citations)
- The Ideation-Execution Gap: Execution Outcomes of LLM-Generated versus Human Research Ideas (6 citations)
- Formalising Human-in-the-Loop: Computational Reductions, Failure Modes, and Legal-Moral Responsibility (3 citations)
- PLANETALIGN: A Comprehensive Python Library for Benchmarking Network Alignment (1 citation)
- SmellNet: A Large-scale Dataset for Real-world Smell Recognition (1 citation)
- The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity? (1 citation)
- What happens when generative AI models train recursively on each others' outputs? (1 citation)
- C-Evolve: Consensus-based Evolution for Prompt Groups
- Control Tax: The Price of Keeping AI in Check
- Counterfactual LLM-based Framework for Measuring Rhetorical Style
- Critical Confabulation: Can LLMs Hallucinate for Social Good?
and 36 more in the interactive map.