mlsanityData to 7 Oct 2026

Test-time compute and efficient thinking

2,144 papers18 this weekLow confidence in this grouping: the papers may belong to more than one topic, or the group may changeLow confidence in this grouping: the papers may belong to more than one topic, or the group may change · About

Keywords: reasoning, test-time, efficient, scaling, llm

Key facts. Test-time compute and efficient thinking: 2,144 papers on arXiv since 2022 (data through 7 Oct 2026). 493 of its 2,144 mapped papers (23%) cite “Tree of Thoughts: Deliberate Problem Solving with Large Language Models” (2023, arXiv 2305.10601). 385 of its 2,144 mapped papers (18%) cite “Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters” (2024, arXiv 2408.03314). 92 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

92 new papers, 17 Sep – 7 Oct; 18 this week.

  1. This week · 18
  2. Invariant-Measure Reasoners: Stable Representations for Latent Reasoning also fits: Reasoning benchmarks and evaluation
  3. SemanticFold: Latent Sequence Compression SeparatesLanguage Modeling, Decodability, and Reasoning
  4. Unrolled Flow Models for Reasoning also fits: Knowledge- and tool-augmented reasoning
  5. Learning Situation-Conditioned Thinking Policies for Long-Term LLM Agents also fits: Reasoning for agents, planning and code
  6. Efficient Reasoning with Flow Language Models also fits: Diffusion reasoning and test-time scaling
  7. The Dichotomy Between Pattern Recognition and Step-by-Step Reasoning also fits: Prompting and chain-of-thought reasoning
  8. A Systematic Study of Small Language Models on Abstract Reasoning Tasks also fits: Reasoning benchmarks and evaluation
  9. Adaptive Power Sampling for LLM Reasoning
  10. ThinkFuse: Trajectory-Aware Test-Time Fusion for Small Reasoning Models
  11. Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation also fits: Reinforcement learning for reasoning
All 92
  1. Trajectory-Retrieval Speculative Decoding: When Does a Model's Own History Help?
  2. Learning What to Distill: Bilevel Top-K Token Selection for Self-Distillation in Large Language Models also fits: Prompting and chain-of-thought reasoning
  3. Learning What to Imitate: Entropy-Aware Distribution Mixing also fits: Diffusion reasoning and test-time scaling
  4. Test-Time Adaptation of Reasoning Strategies with Bayesian Nonparametric Memory also fits: Reasoning benchmarks and evaluation
  5. Quantifying the Stability of Multi-Step Reasoning via Error Amplification also fits: Prompting and chain-of-thought reasoning
  6. DPNL: A DPLL-based Algorithm for Probabilistic Neurosymbolic Learning also fits: Reasoning benchmarks and evaluation
  7. Can Language Models Learn to Reject Their Own Bad Reasoning Steps?
  8. Spend Bytes on Breadth: Precision-Count Trade-offs for Decode-Time KV Compression in Long Chain-of-Thought Reasoning
  9. Earlier · 74
  10. DART-ES: Difficulty-Aware Reweighting and Targeted Replay for Fine-Tuning LLMs with Evolution Strategies also fits: Reinforcement learning for reasoning
  11. Expanding LLM Reasoning
  12. Learning without Overwriting: A Theory of Self-Distillation and Supervised Fine-Tuning in Continual Reasoning
  13. Dynamic Routing as a New Dimension for Test-time Versatility of LLMs also fits: Prompting and chain-of-thought reasoning
  14. Questioning the Questions: Sustaining Self-Evolution in Reasoning Models also fits: Knowledge- and tool-augmented reasoning
  15. SYNLAT: Syntax-Aligned Text-Latent Compression for Chain-of-Thought Reasoning also fits: Prompting and chain-of-thought reasoning
  16. HyperThink: Text-to-Parameter Hypernetworks for Efficient Reasoning
  17. Test-time Calibration Learning for Large Language Model Reasoning also fits: Reinforcement learning for reasoning
  18. From Discovery to Decision: Finite-Budget Recoverability in LLM Voting
  19. ReHoPER: Receding-Horizon Planning for Enhanced Reasoning also fits: Knowledge- and tool-augmented reasoning
  20. Not Too Hard, Not Too Easy: Learning from Intermediate States for LLM Structured Reasoning also fits: Reasoning benchmarks and evaluation
  21. Interpreting Reasoning of Large Language Models via Partial Information Decomposition also fits: Reasoning benchmarks and evaluation
  22. What Limits Recursive Reasoning Models: Optimization, Architecture and Test-Time Scaling
  23. RATIO: Reasoning Analysis and Token-level Inference Optimization for Quantized Reasoning Models
  24. LatentHarness: Learning Latent Actions for Memory and Reasoning via Counterfactual Policy Distillation
  25. Towards Better Exploration in Sequential Test-Time Scaling also fits: Diffusion reasoning and test-time scaling
  26. Taming Speculative Search for Test-Time Scaling in LLM Serving also fits: Mathematical reasoning
  27. CORE: Conflict-Oriented Reasoning Elimination for Verifiable Language-Model Search also fits: Reasoning for agents, planning and code
  28. Adaptive Self-Consistency: From Black-Box Sampling to Distribution-Valued Feedback
  29. Diversity Combining for Multi-Path LLM Reasoning
  30. When Reasoning Goes Astray: Attention Dynamics of Uncontrolled Reasoning also fits: Prompting and chain-of-thought reasoning
  31. StateTree: Enhancing Long-Term Dialogue Reasoning via Reinforcement Learning also fits: Knowledge- and tool-augmented reasoning
  32. Budget Boundary Effects in Test-Time Mathematical Reasoning also fits: Mathematical reasoning
  33. CoEM: Empowering Long-Context Reasoning with Commit-on-Evidence Memory
  34. Provable Test-Time Scaling for Beam Search in LLM Reasoning
  35. Hermes: Learning Contextual Reasoning Unlocks Test-Time Scaling
  36. Explore Broadly, Reason Sharply: Push Small Models toward the Frontier via Sampling also fits: Reinforcement learning for reasoning
  37. Diagnosing and Improving Probabilistic Reasoning in Large Language Models also fits: Reasoning benchmarks and evaluation
  38. S3: Spectral Null-Space Swap Makes Reasoning Models Efficient
  39. Port-Hamiltonian Latent Deliberation: Mitigating the Deliberation Drift Cliff in Test-Time Compute Scaling
  40. MetaCtrl: Your Large Language Models Can Reason Better and More Concisely with a Metacognitive Controller also fits: Reinforcement learning for reasoning
  41. Learning to Prove, Not Just to Answer: Reinforcement Learning from Formal Verification for Natural-Language Logical Reasoning also fits: Reasoning benchmarks and evaluation
  42. Trajectory Soup: Pushing the Compute-Scaling Frontier of LLM Mid-training via Diverse Trajectories
  43. Scheduling Recursive Reasoning in Looped Transformers
  44. When Updating Stops Being Learning: Rethinking LLM Self-Evolution via learnable information gain
  45. Rethinking Reasoning Paths as Phase-Structured Trajectories also fits: Knowledge- and tool-augmented reasoning
  46. Principled Thoughts for Latent Recursive LLM Systems also fits: Reasoning benchmarks and evaluation
  47. Not All Thinking is Created Equal: Latent Reasoning Discovers a Recurrent Search Algorithm for Depth Generalization also fits: Prompting and chain-of-thought reasoning
  48. Continuous Context Management also fits: Reasoning for agents, planning and code
  49. Jailbreaks for Black-Box Uncertainty Quantification in Large Reasoning Models also fits: Reinforcement learning for reasoning
  50. When Confidence Rises Too Early: Detecting Shortcut Reasoning via Premature Answer Commitment also fits: Reasoning benchmarks and evaluation
  51. Making LLMs Truly Forget: Deep Unlearning by Searching, Selecting, and Severing Knowledge Paths also fits: Knowledge- and tool-augmented reasoning
  52. Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge
  53. Test-Time Scaling via Budgeted Multi-Attribute Verification also fits: Reasoning for agents, planning and code
  54. Efficient Reasoning via Constrained Optimization in Latent Space
  55. Steering Language Model Goals with Value Transplant also fits: Reasoning benchmarks and evaluation
  56. Reasoning on the Simplex: Geometric Fixed-Point Models also fits: Mathematical reasoning
  57. DISCO: Distributed Long Context Scaling with Grounding-Reasoning Disaggregation
  58. The Error You See Is Not the Error You Made: Progression-aware Reasoning Origin for Reasoning Error Localization
  59. Calibration, Not Answer Selection: Distilling Internal Confidence in Reasoning Models also fits: Prompting and chain-of-thought reasoning
  60. Where Do Test-Time Scaling and Training Fall Short in Individual Stance Prediction? also fits: Reinforcement learning for reasoning
  61. Beyond the Training Horizon: Mechanisms and Limits of Length Generalization in Looped Transformers also fits: Prompting and chain-of-thought reasoning
  62. Overwhelmed by Choice: Studying LLM Decision Making at Scale also fits: Reasoning benchmarks and evaluation
  63. DualGuard: Dual-Mode Quality Control for Logic-Preserving Data Augmentation also fits: Knowledge- and tool-augmented reasoning
  64. On-Policy Attention Linearization also fits: Multimodal reasoning
  65. Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency
  66. The Price of Thought: Does Test-Time Reasoning Pay in LLM Trading? also fits: Reasoning for agents, planning and code
  67. Neurosymbolic Routing for Reliable Reasoning on Resource-Constrained Edge Devices
  68. Strategic Self-Consistency
  69. SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance
  70. Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs also fits: Diffusion reasoning and test-time scaling
  71. Planned Test-Time Scaling with Coordinated Reasoning Paths
  72. Attention Routing Stabilizes Early: Working-Set Inference for Recurrent Language Models also fits: Diffusion reasoning and test-time scaling
  73. Scaling of Capability and Efficiency at Inference Time in Large Reasoning Models also fits: Prompting and chain-of-thought reasoning
  74. Giving Credit Where It's Due: Redundancy-Aware Learning for Efficient Reasoning also fits: Prompting and chain-of-thought reasoning
  75. Beyond Repeated Sampling: Learning Search Policies for LLM Reasoning
  76. Ladders of Thought: A Self-Evolving Curriculum of Progressively Simplified Reasoning Traces also fits: Prompting and chain-of-thought reasoning
  77. Hill Sampling for Test-Time Scaling: A Simple and Better Alternative to Repeated Sampling, Evolution, and Training also fits: Diffusion reasoning and test-time scaling
  78. LoopCD: Loop-wise Contrastive Decoding for Improving Reasoning in Looped Language Models
  79. Efficient Reasoning Exploration via State-Conditioned Latent Steering with Progress Guidance also fits: Reinforcement learning for reasoning
  80. Measured Joules, Learned Routes: Learning to Route for Energy-Efficient LLM Serving also fits: Reasoning for agents, planning and code
  81. GUARD: Natural Forgetting in Large Reasoning Models via Guided Answer-Reasoning Distillation also fits: Prompting and chain-of-thought reasoning
  82. LogicTrack: Auditing Reasoning Trajectories of Large Language Models with Formal Logic Solvers also fits: Prompting and chain-of-thought reasoning
  83. When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models

Foundations

Most cited by papers here

  1. Tree of Thoughts: Deliberate Problem Solving with Large Language Models2023cited by 493 here
  2. Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters2024cited by 385 here
  3. s1: Simple test-time scaling2025cited by 294 here
  4. Self-Refine: Iterative Refinement with Self-Feedback2023cited by 284 here
  5. Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs2024cited by 270 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 974 here
  2. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 703 here
  3. 2022Large Language Models are Zero-Shot Reasonerscited by 293 here
  4. 2023Let's Verify Step by Stepcited by 485 here
  5. 2023Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotationscited by 187 here
  6. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 430 here
  7. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 207 here
  8. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMscited by 270 here
  9. 2025L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learningcited by 216 here
  10. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Modelscited by 237 here
14 more it builds on, not in this chain
  1. 2017Proximal Policy Optimization Algorithmscited by 184 here
  2. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 284 here
  3. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 493 here
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 175 here
  5. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 209 here
  6. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 385 here
  7. 2024OpenAI o1 System Cardcited by 313 here
  8. 2025O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruningcited by 177 here
  9. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 187 here
  10. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 790 here
  11. 2025s1: Simple test-time scalingcited by 294 here
  12. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMscited by 164 here
  13. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 158 here
  14. 2025Qwen3 Technical Reportcited by 488 here

The 24 most cited of 3,504.

Browse

30 of 2,144, ranked 6–35

  1. 6Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models2025237
  2. 7L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning2025216
  3. 8O1-Pruner: Length-Harmonizing Fine-Tuning for O1-Like Reasoning Pruning2025177
  4. 9TokenSkip: Controllable Chain-of-Thought Compression in LLMs2025164
  5. 10Dynamic Early Exit in Reasoning Models2025147
  6. 11STaR: Bootstrapping Reasoning With Reasoning2022144
  7. 12Token-Budget-Aware LLM Reasoning2024133
  8. 13Language Models (Mostly) Know What They Know2022120
  9. 14ThinkPrune: Pruning Long Chain-of-Thought of LLMs via Reinforcement Learning2025115
  10. 15C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness2024112
All 30
  1. 16CoT-Valve: Length-Compressible Chain-of-Thought Tuning2025111
  2. 17DAST: Difficulty-Adaptive Slow-Thinking for Large Reasoning Models2025108
  3. 18Demystifying Long Chain-of-Thought Reasoning in LLMs2025104
  4. 19Large Language Models Cannot Self-Correct Reasoning Yet2023104
  5. 20Reasoning Models Can Be Effective Without Thinking202598
  6. 21Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models202492
  7. 22gpt-oss-120b & gpt-oss-20b Model Card202591
  8. 23Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs202587
  9. 24Compressed Chain of Thought: Efficient Reasoning Through Dense Representations202484
  10. 25Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning202481
  11. 26AdaptThink: Reasoning Models Can Learn When to Think202577
  12. 27Self-Evaluation Guided Beam Search for Reasoning202376
  13. 28Deep Think with Confidence202575
  14. 29LightThinker: Thinking Step-by-Step Compression202572
  15. 30Training Language Models to Self-Correct via Reinforcement Learning202472
  16. 31Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMs202370
  17. 32Think before you speak: Training Language Models With Pause Tokens202368
  18. 33Self-Training Elicits Concise Reasoning in Large Language Models202566
  19. 34Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach202565
  20. 35Towards Thinking-Optimal Scaling of Test-Time Compute for LLM Reasoning202563