mlsanityData to 14 Oct 2026

AERA: Adaptive Evidence Residual Allocation for Efficient Test-Time Reasoning

arXiv 2608.27964 ↗First posted 28 Aug 2026cs.AIcited by 0

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

Abstract (arXiv)

Test-time scaling improves language-model reasoning by generating additional candidate solutions, but allocating the same inference budget to every problem is computationally wasteful. Existing adaptive stopping methods commonly rely on confidence, agreement, or answer stability, implicitly assuming that stronger current evidence indicates that further computation is unnecessary. We show that this assumption can fail: checkpoint-level correctness evolves non-monotonically, and observable evidence may strengthen before an answer collapses or weaken before it recovers. Motivated by this mismatch, we introduce Adaptive Evidence Residual Allocation (AERA), a sequential controller that learns whether additional computation is likely to recover a better answer from checkpoint-observable evidence. AERA characterizes cumulative response prefixes using answer-distribution, temporal, re-solving, semantic, and compute features, and repeatedly decides whether to stop or allocate the next response block. Future checkpoint correctness is used only to construct offline supervision and is never available to the controller at inference time. Across GSM8K and GPQA Diamond, AERA identifies question-specific residual opportunities while substantially reducing inference computation. In a frozen-threshold incremental-generation evaluation on 300 untouched GSM8K questions, AERA achieves 92.61% accuracy versus 93.01% with 128 responses while reducing completion tokens by 95.99%. These results suggest that adaptive reasoning should estimate the future value of computation rather than equating present confidence with correctness.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 16 of its 34 references that are in the map. First posted on arXiv 28 Aug 2026.

Builds on

16 of 34 references.

Closest ancestors · 11

Same direction, newest first

  1. 2026Reasoning Fails Where Step Flow Breaks8 Apr
  2. 2026SmartThinker: Progressive Chain-of-Thought Length Calibration for Efficient Large Language Model Reasoning9 Mar
  3. 2026Learning from Mistakes: Negative Reasoning Samples Enhance Out-of-Domain Generalization8 Jan
  4. 2025AdaptThink: Reasoning Models Can Learn When to Think19 May
  5. 2025Reasoning Models Can Be Effective Without Thinking14 Apr
  6. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verification7 Apr
  7. 2025Efficient Inference for Large Reasoning Models: A Survey29 Mar
  8. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuning13 Feb
  9. 2025s1: Simple test-time scaling31 Jan
  10. 2024Token-Budget-Aware LLM Reasoning24 Dec
  11. 2024Adaptive Inference-Time Compute: LLMs Can Predict if They Can Do Better, Even Mid-Generation3 Oct

Other references · 5

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 297
  3. 2025ConciseRL: Conciseness-Guided Reinforcement Learning for Efficient Reasoning Modelscited by 14
  4. 2026Re2: Unlocking LLM Reasoning via Reinforcement Learning with Re-solvingcited by 6
  5. 2025AdaNav: Adaptive Reasoning with Uncertainty for Vision-Language Navigationcited by 5

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026BLADE: Boundary-Expanded and Layer-Adaptive Dynamic Exit for Efficient LLM Reasoning
  2. 2026Efficient Test-Time Scaling via Temporal Reasoning Aggregation
  3. 2026Divergent-Convergent Reasoning: Scaling Test-Time Compute through Structured Solution Synthesis
  4. 2026Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models
  5. 2026Interpretable Adaptive Sampling for LLM Test-Time Scaling
  6. 2026ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models
  7. 2025Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering
  8. 2025Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs