mlsanityData to 7 Oct 2026

CoT2-Meta: Budgeted Metacognitive Control for Test-Time Reasoning

arXiv 2603.28135 ↗First posted 30 Mar 2026cs.AIcited by 2

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Recent test-time reasoning methods improve performance by generating more candidate chains or searching over larger reasoning trees, but they typically lack explicit control over when to expand, what to prune, how to repair, and when to abstain. We introduce CoT2-Meta, a training-free metacognitive reasoning framework that combines object-level chain-of-thought generation with meta-level control over partial reasoning trajectories. The framework integrates four components: strategy-conditioned thought generation, tree-structured search, an online process oracle for step-level reasoning evaluation, and a meta-controller that allocates computation through expansion, pruning, repair, stopping, and fallback decisions. Under matched inference budgets, CoT2-Meta consistently outperforms strong single-path, sampling-based, and search-based baselines, including ReST-MCTS. On the default backbone, it achieves 92.8 EM on MATH, 90.4 accuracy on GPQA, 98.65 EM on GSM8K, 75.8 accuracy on BBEH, 85.6 accuracy on MMMU-Pro, and 48.8 accuracy on HLE, with gains over the strongest non-CoT2-Meta baseline of +3.6, +5.2, +1.15, +2.0, +4.3, and +4.3 points, respectively. Beyond these core results, the framework remains effective across a broader 15-benchmark suite spanning knowledge and QA, multi-hop reasoning, coding, and out-of-distribution evaluation. Additional analyses show better compute scaling, improved calibration, stronger selective prediction, targeted repair behavior, and consistent gains across backbone families. These results suggest that explicit metacognitive control is a practical design principle for reliable and compute-efficient test-time reasoning systems.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 48 of its 57 references that are in the map. First posted on arXiv 30 Mar 2026.

Builds on

48 of 57 references.

Closest ancestors · 5

Same direction, newest first

  1. 2026ODAR: Principled Adaptive Routing for LLM Reasoning via Active Inference27 Feb
  2. 2025When To Solve, When To Verify: Compute-Optimal Problem Solving and Generative Verification for LLM Reasoning1 Apr
  3. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  4. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May
  5. 2023Self-Refine: Iterative Refinement with Self-Feedback30 Mar

Other references · 43

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  4. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  5. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,203
  6. 2017On Calibration of Modern Neural Networkscited by 4,860
  7. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  8. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,072
  9. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  10. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  11. 2023Let's Verify Step by Stepcited by 3,598
  12. 2021Program Synthesis with Large Language Modelscited by 3,509
31 more
  1. 2021TruthfulQA: Measuring How Models Mimic Human Falsehoodscited by 2,919
  2. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  3. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  4. 2022Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Themcited by 1,767
  5. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,758
  6. 2022Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Taskscited by 1,057
  7. 2023Graph of Thoughts: Solving Elaborate Problems with Large Language Modelscited by 873
  8. 2022PAL: Program-aided Language Modelscited by 635
  9. 2024MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmarkcited by 494
  10. 2025Process Reinforcement through Implicit Rewardscited by 478
  11. 2020How Can We Know When Language Models Know? On the Calibration of Language Models for Question Answeringcited by 439
  12. 2024ReST-MCTS*: LLM Self-Training via Process Reward Guided Tree Searchcited by 338
  13. 2022Conformal Risk Controlcited by 290
  14. 2025Why Language Models Hallucinatecited by 205
  15. 2025VisualPRM: An Effective Process Reward Model for Multimodal Reasoningcited by 134
  16. 2025BIG-Bench Extra Hardcited by 123
  17. 2025Process Reward Models That Thinkcited by 118
  18. 2024Large Language Models Must Be Taught to Know What They Don't Knowcited by 99
  19. 2025CoT-Kinetics: A Theoretical Modeling Assessing LRM Reasoning Processcited by 50
  20. 2025Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoningcited by 49
  21. 2024Debate on Graph: a Flexible and Reliable Reasoning Framework for Large Language Modelscited by 19
  22. 2025Improving Pre-Trained Vision-Language-Action Policies with Model-Based Searchcited by 12
  23. 2026Beyond Outcome Verification: Verifiable Process Reward Models for Structured Reasoningcited by 12
  24. 2025CEC-Zero: Zero-Supervision Character Error Correction with Self-Generated Rewardscited by 10
  25. 2025Prompt-R1: Collaborative Automatic Prompting Framework via End-to-end Reinforcement Learningcited by 7
  26. 2025Reinforcement Mid-Trainingcited by 6
  27. 2026Are LLM Decisions Faithful to Verbal Confidence?cited by 6
  28. 2026Empirical-MCTS: Continuous Agent Evolution via Dual-Experience Monte Carlo Tree Searchcited by 2
  29. 2025Reasoning in Action: MCTS-Driven Knowledge Retrieval for Large Language Modelscited by 1
  30. 2026TIME: Temporally Intelligent Meta-reasoning Engine for Context-Triggered Explicit Reasoningcited by 1
  31. 2026Knowing When Not to Answer: Abstention-Aware Scientific Reasoningcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Thinking Hard, Not Smart: Reasoning Models Fail to Ration Test-Time Compute Across Questionscited by 1
  2. 2026Know When to Trust the Skill: Delayed Appraisal and Epistemic Vigilance for Single-Agent LLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought
  2. 2025Meta-R1: Empowering Large Reasoning Models with Metacognition
  3. 2025Adapting Like Humans: A Metacognitive Agent with Test-time Reasoning
  4. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time
  5. 2026Think2: Grounded Metacognitive Reasoning in Large Language Models
  6. 2026LLMs Know When They Know, but Do Not Act on It: A Metacognitive Harness for Test-time Scaling
  7. 2025Monitor-Generate-Verify (MGV): Formalising Metacognitive Theory for Language Model Reasoning
  8. 2026Beyond Meta-Reasoning: Metacognitive Consolidation for Self-Improving LLM Reasoning