mlsanityData to 14 Oct 2026

Inference Time Optimization with Confidence Dynamics

arXiv 2605.25244 ↗First posted 24 May 2026cs.CLcited by 2

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

Abstract (arXiv)

Inference time optimization techniques, such as repeated sampling, have significantly advanced the reasoning capabilities of Large Language Models (LLMs). However, the critical role of model uncertainty remains largely underexplored in these optimization strategies. In this paper, we investigate the dynamics of confidence along reasoning trajectories and for first time reveal a surprising and unique pattern: correct answer traces tend to exhibit confidence improvement over time (positive confidence gain), while incorrect traces show attenuated or declining confidence as reasoning proceeds. Based on this observation, we propose Confidence Dynamic Gain (CDG) based voting, which incorporates how the confidence trajectory of the response evolves along the reasoning chain. Experiments across four open-source architectures (DeepSeek-R1, gpt-oss, Gemma-3, Qwen-QwQ) on the AIME24/25, HMMT25, and BRUMO25 benchmarks demonstrate that CDG yields a significant performance boost over baselines. These results demonstrate that our method provides a robust discriminative signal for improving answer selection in LLM reasoning. We also provide theoretical insights for this phenomenon. Code will be released at https://github.com/Accenture/CDG.git.

Key facts. Cited by 2 papers 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 26 of its 31 references that are in the map. First posted on arXiv 24 May 2026.

Builds on

26 of 31 references.

Closest ancestors · 5

Same direction, newest first

  1. 2025Deep Think with Confidence21 Aug
  2. 2025gpt-oss-120b & gpt-oss-20b Model Card8 Aug
  3. 2025Ranked Voting based Self-Consistency of Large Language Models16 May
  4. 2025Scalable Best-of-N Selection for Large Language Models via Self-Certainty25 Feb
  5. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug

Other references · 21

  1. 2017Attention Is All You Needcited by 60,361
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  5. 2025Qwen3 Technical Reportcited by 8,820
  6. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  7. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,703
  8. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,455
  9. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,879
  10. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  11. 2018Self-Attention with Relative Position Representationscited by 1,120
  12. 2022Solving math word problems with process- and outcome-based feedbackcited by 901
9 more
  1. 2025MathArena: Evaluating LLMs on Uncontaminated Math Competitionscited by 297
  2. 2025Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Freecited by 283
  3. 2025See What You Are Told: Visual Attention Sink in Large Multimodal Modelscited by 132
  4. 2023Self-Evaluation Improves Selective Generation in Large Language Modelscited by 76
  5. 2024Learning to Route LLMs with Confidence Tokenscited by 67
  6. 2025Seeing Far and Clearly: Mitigating Hallucinations in MLLMs with Attention Causal Decodingcited by 43
  7. 2025KVSink: Understanding and Enhancing the Preservation of Attention Sinks in KV Cache Quantization for LLMscited by 20
  8. 2025MCA-LLaVA: Manhattan Causal Attention for Reducing Hallucination in Large Vision-Language Modelscited by 12
  9. 2025Mind the Generation Process: Fine-Grained Confidence Estimation During LLM Generationcited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026TrAC: Trace-Conditioned Answer Consistency for Efficient Uncertainty Quantification in LLMscited by 0
  2. 2026Consilience for Verifier-Free Test-Time Scalingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Confidence over Time: Confidence Calibration with Temporal Logic for Large Language Model Reasoning
  2. 2026Early Stopping for Large Reasoning Models via Confidence Dynamics
  3. 2026Beyond Confidence: Stability-Aware Test-Time Adaptation for LLM Reasoning
  4. 2025Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering
  5. 2026Recurrent Confidence Chain: Temporal-Aware Uncertainty Quantification in Large Language Models
  6. 2026Future Confidence Distillation in Large Language Models
  7. 2025Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs
  8. 2026Confidence Before Answering: A Paradigm Shift for Efficient LLM Uncertainty Estimation