mlsanityData to 7 Oct 2026

Think in Parallel, Answer as One: Logit Averaging for Open-Ended Reasoning

arXiv 2512.02874 ↗First posted 2 Dec 2025cs.CLcited by 5

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

Abstract (arXiv)

Majority voting has proven effective for close-ended question answering by aggregating parallel reasoning traces. However, it is not directly applicable to open-ended reasoning, such as code generation and web-based deep research, where a "majority" over complete solutions is ill-defined. We introduce ThinkMerge, a training-free, plug-and-play decoding strategy that runs K parallel reasoning traces and averages their next-token logits at synchronization points to produce a single coherent output. ThinkMerge integrates seamlessly with vLLM/SGLang and remains compatible with standard decoding techniques such as Top-p/Top-k. Empirically, it matches or surpasses majority voting on AIME and GPQA, while delivering consistent gains on open-ended coding tasks: on LiveCodeBench (hard), pass@1 improves by +8.28% for DeepCoder-14B-Preview and +7.58% for Qwen3-8B. Beyond code, we further show that ThinkMerge improves web-based deep-research agents (e.g., WebSailor-7B/32B) across GAIA, BrowseComp-en/zh, and XbenchDeepSearch. These results demonstrate that parallel test-time scaling can benefit open-ended reasoning without relying on voting over complete outputs.

Key facts. Cited by 5 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 43 of its 58 references that are in the map. First posted on arXiv 2 Dec 2025.

Builds on

43 of 58 references.

Closest ancestors · 12

Same direction, newest first

  1. 2025Learning to Reason Across Parallel Samples for LLM Reasoning10 Jun
  2. 2025Wait, We Don't Need to "Wait"! Removing Thinking Tokens Improves Reasoning Efficiency10 Jun
  3. 2025Understanding Aha Moments: from External Observations to Internal Mechanisms3 Apr
  4. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  5. 2025Revisiting the Test-Time Scaling of o1-like Models: Do they Truly Possess Test-Time Scaling Capabilities?17 Feb
  6. 2025The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks12 Feb
  7. 2025Can 1B LLM Surpass 405B LLM? Rethinking Compute-Optimal Test-Time Scaling10 Feb
  8. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
  9. 2024Mirror-Consistency: Harnessing Inconsistency in Majority Voting7 Oct
  10. 2024Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models1 Aug
  11. 2023Universal Self-Consistency for Large Language Model Generation29 Nov
  12. 2023Let's Sample Step by Step: Adaptive-Consistency for Efficient Reasoning and Coding with LLMs19 May

Other references · 31

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  5. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  6. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  7. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  8. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  9. 2024From Local to Global: A Graph RAG Approach to Query-Focused Summarizationcited by 1,411
  10. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,107
  11. 2023GAIA: a benchmark for General AI Assistantscited by 1,084
  12. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 946
19 more
  1. 2025BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agentscited by 616
  2. 2023LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusioncited by 550
  3. 2025WebThinker: Empowering Large Reasoning Models with Deep Research Capabilitycited by 320
  4. 2025MathArena: Evaluating LLMs on Uncontaminated Math Competitionscited by 275
  5. 2025WebSailor: Navigating Super-human Reasoning for Web Agentcited by 218
  6. 2024SWE-bench Multimodal: Do AI Systems Generalize to Visual Software Domains?cited by 196
  7. 2025When More is Less: Understanding Chain-of-Thought Length in LLMscited by 182
  8. 2025BrowseComp-ZH: Benchmarking Web Browsing Ability of Large Language Models in Chinesecited by 139
  9. 2024BoNBoN Alignment for Large Language Models and the Sweetness of Best-of-n Samplingcited by 120
  10. 2025xbench: Tracking Agents Productivity Scaling with Profession-Aligned Real-World Evaluationscited by 118
  11. 2024A Thorough Examination of Decoding Methods in the Era of LLMscited by 112
  12. 2024Fast Best-of-N Decoding via Speculative Rejectioncited by 112
  13. 2025An Empirical Study on Eliciting and Improving R1-like Reasoning Modelscited by 78
  14. 2024The Limits of Inference Scaling Through Resamplingcited by 51
  15. 2025The Majority is not always right: RL training for solution aggregationcited by 46
  16. 2024Bridging the Gap between Different Vocabularies for LLM Ensemblecited by 44
  17. 2025LLMs Can Generate a Better Answer by Aggregating Their Own Responsescited by 27
  18. 2024M-Ped: Multi-Prompt Ensemble Decoding for Large Language Modelscited by 3
  19. 2025Token-level Ensembling of Models with Different Vocabulariescited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026StAR: Segment Anything Reasonercited by 2
  2. 2026Intrinsic Selection and Particle Resampling for Inference-Time Scaling Beyond Domain Verifiabilitycited by 1
  3. 2026Hyperdimensional Cross-Modal Alignment of Frozen Language and Image Models for Efficient Image Captioningcited by 0
  4. 2026CIPHER: A Decoupled Exploration-Selection Framework for Test-Time Scaling of Data Science Agentscited by 0
  5. 2026Self-Repulsive Sampling for Diffusion Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding
  2. 2026SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation
  3. 2025Logit Arithmetic Elicits Long Reasoning Capabilities Without Training
  4. 2025First Finish Search: Efficient Test-Time Scaling in Large Language Models
  5. 2023Skeleton-of-Thought: Prompting LLMs for Efficient Parallel Generation
  6. 2025ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models
  7. 2025DeepPrune: Parallel Scaling without Inter-trace Redundancy
  8. 2025Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts