ICML 2026 accepted papers on Large language models
1,159 of the 6,341 papers accepted at ICML 2026 (18.3%) are on Large language models, compared with 16.1% at ICML 2025. They fall into 7 topics.
See ICML 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
Language, safety and interpretability · 314 papers
- Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units (7 citations)
- mmBERT: A Modern Multilingual Encoder with Annealed Language Learning (4 citations)
- The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models (4 citations)
- Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity (2 citations)
- A model of errors in transformers (1 citation)
- MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety (1 citation)
- A Benchmark and Framework for Evaluating Next Action Predictions in Spreadsheets
- A geometric relation of the error introduced by sampling a language model's output distribution to its internal state
- A Judge-Aware Ranking Framework for Evaluating Large Language Models without Ground Truth
- A Unified Approach to Interpreting Knowledge Distillation for Large Language Models via Interactions
- ABCD: All Biases Come Disguised
- Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers
and 302 more in the interactive map.
Reasoning and chain of thought · 306 papers
- Chain-of-Thought Reasoning In The Wild Is Not Always Faithful (2 citations)
- ConServe: Fine-Grained GPU Harvesting for LLM Online and Offline Co-Serving (1 citation)
- HybridFlow: Resource-Adaptive Subtask Routing for Efficient Edge-Cloud LLM Inference (1 citation)
- Learning to Discover at Test Time (1 citation)
- Monitoring Monitorability (1 citation)
- Optimal Self-Consistency for Efficient Reasoning with Large Language Models (1 citation)
- A Queueing-Theoretic Framework for Stability Analysis of LLM Inference with KV Cache Memory Constraints
- A Task-centric Theory for Iterative Self-Improvement with Easy-to-Hard Curricula
- Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning
- Adaptive Testing for LLM Evaluation: A Psychometric Alternative to Static Benchmarks
- Adversarial Training for Process Reward Models
- AlgoTrace: Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
and 294 more in the interactive map.
Agents, code and math · 237 papers
- BrokenMath: A Benchmark for Sycophancy in Theorem Proving with LLMs (3 citations)
- DeepAnalyze: Agentic Large Language Models for Autonomous Data Science (3 citations)
- SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks? (2 citations)
- ACON: Optimizing Context Compression for Long-horizon LLM Agents (1 citation)
- AICrypto: Evaluating Cryptography Capabilities of Large Language Models (1 citation)
- DEER: A Benchmark for Evaluating Deep Research Agents on Expert Report Generation (1 citation)
- Efficient Distributed MLLM Training with Cornstarch (1 citation)
- Meta Context Engineering via Agentic Skill Evolution (1 citation)
- On Path to Multimodal Historical Reasoning: HistBench and HistAgent (1 citation)
- PaperBanana: Automating Academic Illustration for AI Scientists (1 citation)
- Quantifying Frontier LLM Capabilities for Container Sandbox Escape (1 citation)
- Retro-Expert: Collaborative Reasoning for Interpretable Retrosynthesis (1 citation)
and 225 more in the interactive map.
Instruction and fine-tuning · 114 papers
- Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning (1 citation)
- LIFT: A Novel Framework for Enhancing Long-Context Understanding of LLMs via Long Input Fine-Tuning (1 citation)
- Probability-Entropy Calibration: An Elastic Indicator for Adaptive Fine-tuning (1 citation)
- $f$-Divergence Self-Play for Tabular Anomaly Detection via Large Language Models
- 3D-RFT: Reinforcement Fine-Tuning for Video-based 3D Scene Understanding
- A Critical Look at Targeted Instruction Selection: Disentangling What Matters (and What Doesn’t)
- A Language-Guided Bayesian Optimization for Efficient LoRA Hyperparameter Search
- AdaMeZO: Adam-style Zeroth-Order Optimizer for LLM Fine-tuning Without Maintaining the Moments
- AIR: Post-training Data Selection for Reasoning via Attention Head Influence
- AlienLM: Alienization of Language for API-Boundary Privacy in Black-Box LLMs
- BARRED: Synthetic Training of Custom Policy Guardrails via Asymmetric Debate
- Beyond Log Likelihood: Probability-Based Objectives for Supervised Fine-Tuning across the Model Capability Continuum
and 102 more in the interactive map.
Alignment and preferences · 76 papers
- LILO: Bayesian Optimization with Natural Language Feedback (2 citations)
- A Regret Minimization Framework on Preference Learning in Large Language Models
- ActiveUltraFeedback: Efficient Preference Data Generation using Active Learning
- Alignment Tampering: How Reinforcement Learning from Human Feedback Is Exploited to Optimize Misaligned Biases
- Asymptotic Universal Alignment: A New Alignment Framework via Test-Time Scaling
- Autoregressive Direct Preference Optimization
- CAMEL: Confidence-Gated Reflection for Reward Modeling
- Causal Direct Preference Optimization for Distributionally Robust Generative Recommendation
- Closing the Expression Gap in LLM Instructions via Socratic Questioning
- Configurable Reward Model for Balanced Safety Alignment
- Contrastive Weak-to-Strong Generalization
- Controllable and Explainable Personality Sliders for LLMs at Inference Time
and 64 more in the interactive map.
Retrieval and knowledge · 73 papers
- MultiHal: Multilingual Dataset for Knowledge-Graph Grounded Evaluation of LLM Hallucinations (1 citation)
- A Geometric Analysis of Small-sized Language Model Hallucinations
- BESPOKE: Benchmark for Search-Augmented Large Language Model Personalization via Diagnostic Feedback
- Bridging the Knowledge-Prediction Gap in LLMs on Multiple-Choice Questions
- Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts
- DTKG: Dual-Track Knowledge Graph-Verified Reasoning Framework for Multi-Hop QA
- Empty Shelves or Lost Keys? Recall Is the Bottleneck for Parametric Factuality
- Evidential Reasoning Advances Interpretable Real-World Disease Screening
- ExpWeaver: LLM Agents Learn from Experience via Latent RAG
- Failure is Feedback: History-Aware Backtracking for Agentic Traversal in Multimodal Graphs
- From Conflict to Consensus: Boosting Medical Reasoning via Multi-Round Agentic RAG
- From Volume to Value: Preference-Aligned Memory Construction for On-Device RAG
and 61 more in the interactive map.
AI and society · 39 papers
- When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation (2 citations)
- Who’s in Charge? Disempowerment Patterns in Real-World LLM Usage (2 citations)
- IDRBench: Understanding the Capability of Large Language Models on Interdisciplinary Research (1 citation)
- Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations (1 citation)
- AgentExpt: Automating AI Experiment Design with LLM-based Resource Retrieval Agent
- AI Cartography: Mapping the Latent Landscape of AI Benchmark Ecosystems
- Alignment Pretraining: AI Discourse Causes Self-Fulfilling (Mis)alignment
- BRIDGE: Predicting Human Task Completion Time From Model Performance
- CentaurEval: Benchmarking Human-in-the-Loop Value in Agentic Coding
- CoCoReviewBench: A Completeness- and Correctness-Oriented Benchmark for AI Reviewers
- Computational Arbitrage in AI Model Markets
- Conf-Gen: Conformal Uncertainty Quantification for Generative Models
and 27 more in the interactive map.