mlsanityData to 7 Oct 2026

Efficient and Trainable Language Model Test-Time Scaling via Local Branch Routing

arXiv 2606.25354 ↗First posted 24 Jun 2026cs.CLcited by 1

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

Abstract (arXiv)

Test-time scaling improves language-model reasoning, but existing approaches often face a difficult trade-off: long chain-of-thought sampling remains single-threaded, while sentence- or solution-level search can be computationally expensive and hard to train end-to-end. We introduce Local Branch Routing (LBR), a token-level test-time scaling framework that expands a small local lookahead tree, forwards all sampled branches through the language model, and uses a lightweight router to select the depth-1 subtree to commit. By routing over the hidden states of candidate local futures, LBR allows each token decision to use evidence beyond the root next-token distribution while avoiding full solution-level search. The resulting prune-shift-grow decoding process preserves discrete branch identities and defines a tractable tree-trajectory likelihood: newly grown nodes are counted when first sampled, and router decisions are assigned explicit probabilities. This enables end-to-end reinforcement learning with verifiable rewards, jointly optimizing the base model and router under the same likelihood-ratio principle as discrete-token RLVR. On synthetic hierarchical-planning tasks, LBR shows that post-candidate hidden states provide useful routing evidence. On mathematical reasoning benchmarks, LBR improves both Pass@1 and Pass@32 over discrete chain-of-thought, vanilla discrete-token RLVR, and RL-compatible soft-token branching baselines. These results suggest that lightweight local branching offers an efficient, trainable, and discrete form of language-model test-time scaling.

Key facts. Cited by 1 paper 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 31 of its 34 references that are in the map. First posted on arXiv 24 Jun 2026.

Builds on

31 of 34 references.

Closest ancestors · 8

Same direction, newest first

  1. 2025Scaling Speculative Decoding with Lookahead Reasoning24 Jun
  2. 2025Efficient Reasoning for LLMs through Speculative Chain-of-Thought27 Apr
  3. 2025Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Branching Tree Search6 Mar
  4. 2025Policy Guided Tree Search for Enhanced LLM Reasoning4 Feb
  5. 2025s1: Simple test-time scaling31 Jan
  6. 2025Reward-Guided Speculative Decoding for Efficient LLM Reasoning31 Jan
  7. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
  8. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May

Other references · 23

  1. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  4. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
  5. 2016Understanding intermediate layers using linear classifier probescited by 1,279
  6. 2023Accelerating Large Language Model Decoding with Speculative Samplingcited by 975
  7. 2023Reasoning with Language Model is Planning with World Modelcited by 846
  8. 2021Probing Classifiers: Promises, Shortcomings, and Advancescited by 831
  9. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  10. 2023The Internal State of an LLM Knows When It's Lyingcited by 644
  11. 2019Designing and Interpreting Probes with Control Taskscited by 608
  12. 2023Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Modelscited by 551
11 more
  1. 2024Break the Sequential Dependency of LLM Inference Using Lookahead Decodingcited by 295
  2. 2024Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learningcited by 238
  3. 2025Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Spacecited by 109
  4. 2025TreeRL: LLM Reinforcement Learning with On-Policy Tree Searchcited by 75
  5. 2025Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thoughtcited by 73
  6. 2025TreePO: Bridging the Gap of Policy Optimization and Efficacy and Inference Efficiency with Heuristic Tree-based Modelingcited by 60
  7. 2025LLMs are Single-threaded Reasoners: Demystifying the Working Mechanism of Soft Thinkingcited by 34
  8. 2025Continuous Chain of Thought Enables Parallel Exploration and Reasoningcited by 32
  9. 2025Soft Tokens, Hard Truthscited by 31
  10. 2026Multiplex Thinking: Reasoning via Token-wise Branch-and-Mergecited by 16
  11. 2025DeepSearch: Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Searchcited by 11

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026FutureBridge: Token Selection Beyond Local Preference in Collaborative Decodingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025HierRouter: Coordinated Routing of Specialized Large Language Models via Reinforcement Learning
  2. 2026When LLM Meets Tree Search: A Systematic View of Inference as Search in Large Language Models
  3. 2025Chain-in-Tree: Back to Sequential Reasoning in LLM Tree Search
  4. 2026Towards Better Exploration in Sequential Test-Time Scaling
  5. 2025Test-time Scaling of LLMs: A Survey from A Subproblem Structure Perspective
  6. 2025Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models
  7. 2025T1: Advancing Language Model Reasoning through Reinforcement Learning and Inference Scaling
  8. 2026Policy-Guided Stepwise Model Routing for Cost-Effective Reasoning