mlsanityData to 7 Oct 2026

Towards Better Exploration in Sequential Test-Time Scaling

arXiv 2609.39632 ↗First posted 30 Sep 2026cs.LGcited by 0

On the map: LLM reasoning → Test-time compute and efficient thinking also fits: Diffusion reasoning and test-time scaling

Abstract (arXiv)

Test-time scaling improves language model reasoning by spending additional compute at inference. However, both classes of existing methods often fail to continue improving over long timescales. Parallel methods repeatedly sample independent answers from the model, scaling poorly on problems the model is unlikely to solve in a single attempt. In contrast, sequential methods build on previous answers to access new ideas, yet so far have not been shown to reach answers beyond those found by parallel scaling. First, we show that sequential scaling often stops improving because it becomes prematurely trapped in an attractor: a set of answers that prevents exploration of different answers once entered. Across 27 combinations of scaling methods, models, and benchmarks, we find that 53.8% of sequential scaling trajectories enter an attractor within four iterations. Second, we show that a simple model-mixing intervention helps escape attractors. This reduces the attractor hit rate by 21.2 percentage points on average, expands solution coverage beyond a compute-matched parallel baseline, and improves accuracy of recursive self-aggregation by at least 2.2 percentage points. Our results motivate refocusing long-horizon test-time scaling from parallel methods to sequential methods that improve previous answers.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 59 of its 67 references that are in the map. First posted on arXiv 30 Sep 2026.

Builds on

59 of 67 references.

Closest ancestors · 19

Same direction, newest first

  1. 2026Reasoning as Attractor Dynamics: Latent Memory Retrieval via Gibbs-Weighted Energy Minimization23 Jun
  2. 2026Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning20 May
  3. 2026Solve the Loop: Attractor Models for Language and Reasoning12 May
  4. 2025Recursive Language Models31 Dec
  5. 2025Geometric Dynamics of Agentic Loops in Large Language Models11 Dec
  6. 2025Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models30 Sep
  7. 2025Scaling over Scaling: Exploring Test-Time Scaling Plateau in Large Reasoning Models26 May
  8. 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?31 Mar
  9. 2025InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models9 Mar
  10. 2025Revisiting the Test-Time Scaling of o1-like Models: Do they Truly Possess Test-Time Scaling Capabilities?17 Feb
  11. 2025Rethinking Mixture-of-Agents: Is Mixing Different Large Language Models Beneficial?2 Feb
  12. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
7 more
  1. 2024Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models1 Aug
  2. 2024On the Self-Verification Limitations of Large Language Models on Reasoning and Planning Tasks12 Feb
  3. 2023Think before you speak: Training Language Models With Pause Tokens3 Oct
  4. 2023Large Language Models Cannot Self-Correct Reasoning Yet3 Oct
  5. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May
  6. 2023Self-Refine: Iterative Refinement with Self-Feedback30 Mar
  7. 2022STaR: Bootstrapping Reasoning With Reasoning28 Mar

Other references · 40

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  3. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
  4. 2023Let's Verify Step by Stepcited by 3,598
  5. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  6. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  7. 2023Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debatecited by 995
  8. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 948
  9. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 946
  10. 2025AlphaEvolve: A coding agent for scientific and algorithmic discoverycited by 763
  11. 2018Universal Transformerscited by 613
  12. 2023LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusioncited by 550
28 more
  1. 2025GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learningcited by 449
  2. 2024Mixture-of-Agents Enhances Large Language Model Capabilitiescited by 420
  3. 2025Accelerating scientific discovery with Co-Scientistcited by 353
  4. 2023ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMscited by 297
  5. 2024When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMscited by 230
  6. 2024The Effect of Sampling Temperature on Problem Solving in Large Language Modelscited by 168
  7. 2024Recursive Introspection: Teaching Language Model Agents How to Self-Improvecited by 162
  8. 2025REASONING GYM: Reasoning Environments for Reinforcement Learning with Verifiable Rewardscited by 121
  9. 2025Multi-Agent Design: Optimizing Agents with Better Prompts and Topologiescited by 106
  10. 2026Learning to Discover at Test Timecited by 94
  11. 2025Debate or Vote: Which Yields Better Decisions in Multi-Agent Large Language Models?cited by 74
  12. 2025Towards Robust Mathematical Reasoningcited by 67
  13. 2025Stop Overvaluing Multi-Agent Debate -- We Must Rethink Evaluation and Embrace Model Heterogeneitycited by 43
  14. 2026From Entropy to Epiplexity: Rethinking Information for Computationally Bounded Intelligencecited by 39
  15. 2025AMO-Bench: Large Language Models Still Struggle in High School Math Competitionscited by 36
  16. 2026Understanding Agent Scaling in LLM-Based Multi-Agent Systems via Diversitycited by 25
  17. 2026Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loopscited by 25
  18. 2025Unveiling Attractor Cycles in Large Language Models: A Dynamical Systems View of Successive Paraphrasingcited by 14
  19. 2024Mitigating Copy Bias in In-Context Learning through Neuron Pruningcited by 11
  20. 2026Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RLcited by 11
  21. 2025Multi-LLM Collaborative Search for Complex Problem Solvingcited by 9
  22. 2026Dream-RSI: Recursive Self-Improvement through Evolving Worldscited by 9
  23. 2026Diversity Collapse in Multi-Agent LLM Systems: Structural Coupling and Collective Failure in Open-Ended Idea Generationcited by 7
  24. 2025MiGrATe: Mixed-Policy GRPO for Adaptation at Test-Timecited by 6
  25. 2026Attractor States Emerge in Multi-Turn LLM Conversationscited by 6
  26. 2025Generalizing Test-time Compute-optimal Scaling as an Optimizable Graphcited by 4
  27. 2026Mutation Without Variation: Convergence Dynamics in LLM-Driven Program Evolutioncited by 4
  28. 2026Do Language Models Converge to Themselves? Recursive Self-Refinement as Textual Relaxationcited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025The Art of Scaling Test-Time Compute for Large Language Models
  2. 2026Test-Time Scaling in the Wild: Why Exploitation, Not Exploration, Is the Bottleneck
  3. 2025The Sequential Edge: Inverse-Entropy Voting Beats Parallel Self-Consistency at Matched Compute
  4. 2026Test-Time Scaling in Reasoning LLMs: Inference Regimes, Evaluation, and Reproducibility
  5. 2026Efficient Test-Time Scaling via Temporal Reasoning Aggregation
  6. 2025Let Me Think! A Long Chain-of-Thought Can Be Worth Exponentially Many Short Ones
  7. 2025On the Role of Temperature Sampling in Test-Time Scaling
  8. 2025Are We Scaling the Right Thing? A System Perspective on Test-Time Scaling