mlsanityData to 14 Oct 2026

On the Overscaling Curse of Parallel Thinking: System Efficacy Contradicts Sample Efficiency

arXiv 2601.21619 ↗First posted 29 Jan 2026cs.LG cs.AI cs.CLcited by 1

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

Abstract (arXiv)

Parallel thinking improves LLM reasoning through multi-path sampling and aggregation. In standard evaluations, due to a lack of sample-specific priors, all samples share a global budget chosen to maximize dataset accuracy. However, many samples reach their best accuracy with much smaller budgets, causing low budget utilization. This contradiction between system efficacy and sample efficiency constitutes the Overscaling Curse. In this paper, we first provide a formal analysis of the overscaling curse and quantify its prevalence and severity in real-world systems. To break it, we propose Latent Budget Predictor (LanBo), which probes model latent representations to predict sample-specific optimal budgets. LanBo significantly improves budget utilization while maintaining dataset accuracy. We further integrate LanBo into the full decoding pipeline, inspiring Pre-decoding Budget Adaptation (PreAda), a paradigm that allocates budgets before decoding to preserve decoding-time parallelization. LanBo substantially improves hardware-aware efficiency in latency and memory, demonstrating both its practical value and the promise of LanBo for efficient parallel decoding.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 63 of its 67 references that are in the map. First posted on arXiv 29 Jan 2026.

Builds on

63 of 67 references.

Closest ancestors · 17

Same direction, newest first

  1. 2025Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning8 Dec
  2. 2025Pushing Test-Time Scaling Limits of Deep Search with Asymmetric Verification7 Oct
  3. 2025Slim-SC: Thought Pruning for Efficient Scaling with Self-Consistency17 Sep
  4. 2025Parallel-R1: Towards Parallel Thinking via Reinforcement Learning9 Sep
  5. 2025ParaThinker: Native Parallel Thinking as a New Paradigm to Scale LLM Test-time Compute30 Aug
  6. 2025Deep Think with Confidence21 Aug
  7. 2025Parallel Scaling Law for Language Models15 May
  8. 2025Sampling-Efficient Test-Time Scaling: Self-Estimating the Best-of-N Sampling in Early Decoding3 Mar
  9. 2025Scalable Best-of-N Selection for Large Language Models via Self-Certainty25 Feb
  10. 2025Dynamic Parallel Tree Search for Efficient LLM Reasoning22 Feb
  11. 2025s1: Simple test-time scaling31 Jan
  12. 2024Make Every Penny Count: Difficulty-Adaptive Self-Consistency for Cost-Efficient Reasoning24 Aug
5 more
  1. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  2. 2024Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning19 Jan
  3. 2023Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation28 Jul
  4. 2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMs19 May
  5. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May

Other references · 46

  1. 2017Attention Is All You Needcited by 60,361
  2. 2020Language Models are Few-Shot Learnerscited by 32,419
  3. 2023GPT-4 Technical Reportcited by 18,396
  4. 2024The Llama 3 Herd of Modelscited by 14,715
  5. 2015Distilling the Knowledge in a Neural Networkcited by 10,324
  6. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  7. 2025Qwen3 Technical Reportcited by 8,820
  8. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  9. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  10. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  11. 2024Qwen2.5 Technical Reportcited by 4,942
  12. 2023Let's Verify Step by Stepcited by 3,789
34 more
  1. 2019The Curious Case of Neural Text Degenerationcited by 2,916
  2. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  3. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,810
  4. 2024MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmarkcited by 1,848
  5. 2018Hierarchical Neural Story Generationcited by 1,193
  6. 2023Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debatecited by 1,011
  7. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 991
  8. 2025OpenAI GPT-5 System Cardcited by 989
  9. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
  10. 2023The Internal State of an LLM Knows When It's Lyingcited by 659
  11. 2025BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agentscited by 647
  12. 2023LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusioncited by 564
  13. 2025LIMO: Less is More for Reasoningcited by 498
  14. 2025Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRscited by 377
  15. 2024Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Expertscited by 376
  16. 2025rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinkingcited by 313
  17. 2025Layer by Layer: Uncovering Hidden Representations in Language Modelscited by 311
  18. 2023Improving Text Embeddings with Large Language Modelscited by 303
  19. 2025DeepMath-103K: A Large-Scale, Challenging, Decontaminated, and Verifiable Mathematical Dataset for Advancing Reasoningcited by 280
  20. 2025Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergycited by 214
  21. 2024Accessing GPT-4 level Mathematical Olympiad Solutions via Monte Carlo Tree Self-refine with LLaMa-3 8Bcited by 158
  22. 2025BrowseComp-ZH: Benchmarking Web Browsing Ability of Large Language Models in Chinesecited by 145
  23. 2025General-Reasoner: Advancing LLM Reasoning Across All Domainscited by 135
  24. 2016Sequence-to-Sequence Learning as Beam-Search Optimizationcited by 127
  25. 2024Fast Best-of-N Decoding via Speculative Rejectioncited by 112
  26. 2025TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modelingcited by 61
  27. 2024Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluationcited by 57
  28. 2025PolyMath: Evaluating Mathematical Reasoning in Multilingual Contextscited by 56
  29. 2025Hogwild! Inference: Parallel LLM Generation via Concurrent Attentioncited by 32
  30. 2025Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generationcited by 30
  31. 2025Emergent Response Planning in LLMscited by 19
  32. 2024Embedding Trajectory for Out-of-Distribution Detection in Mathematical Reasoningcited by 18
  33. 2025ParallelMuse: Agentic Parallel Thinking for Deep Information Seekingcited by 14
  34. 2025Generalized Parallel Scaling with Interdependent Generationscited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Group-Marginalized Self-Rewarding RL Drives Zero-Label Self-Evolvingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Faster and Better LLMs via Latency-Aware Test-Time Scaling
  2. 2025Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling
  3. 2026Parason: Revealing Subtask and Trial Parallelism in LLM Reasoning
  4. 2024Cerberus: Efficient Inference with Adaptive Parallel Decoding and Sequential Knowledge Enhancement
  5. 2025ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models
  6. 2025ParallelBench: Understanding the Trade-offs of Parallel Decoding in Diffusion LLMs
  7. 2025Overclocking LLM Reasoning: Monitoring and Controlling Thinking Path Lengths in LLMs
  8. 2026Characterizing Parallelism Strategies in LLM Inference: Fundamental Compute-Communication Trade-offs