DeepLook: Uncertainty-Guided Branching with Lookahead Confidence for Efficient Reasoning

Tingxin Yang; Zefeng Wang; Mengyue Wang; Jinhe Bi; Xingcheng Zhou; Volker Tresp; Yunpu Ma

cs.AIJun 12, 2026

DeepLook: Uncertainty-Guided Branching with Lookahead Confidence for Efficient Reasoning

Authors: Tingxin Yang, Zefeng Wang, Mengyue Wang, Jinhe Bi, Xingcheng Zhou, Volker Tresp, Yunpu Ma

Organizations: LMU Munich Munich, Germany · MCML, LMU, MemAgents Lab Munich, Germany

Abstract

Test-time scaling improves large language model reasoning, but spends compute uniformly along a reasoning trace, even in stretches that are already stable. Motivated by the observation that reasoning failures show an early onset of uncertainty, we introduce DeepLook, a training-free decoding framework that monitors segment-level confidence, branches only at uncertainty bottlenecks, and ranks candidate continuations by Average Lookahead Confidence (ALC)---their mean confidence over fixed-horizon rollouts---before completing and voting over a pruned subset. On four competition mathematics benchmarks across four open models, DeepLook matches or exceeds the accuracy of DeepConf-low in 12 of 16 settings while generating 87.3% fewer tokens, showing that a model's own confidence is informative enough to decide where extra decoding compute is worth spending. On knowledge-intensive GPQA-Diamond the accuracy advantage disappears even though the token saving does not: coherent hallucinations also receive high ALC, so internal confidence localizes instability, not falsehood. Code is available at https://github.com/txyang1/deeplook.

Figures & tables

Figure 1: DeepLook conceptual overview and accuracy–token-cost frontier. Left: uniform brute-force scaling vs. selective lookahead at uncertainty bottlenecks. Middle: the DeepLook pipeline of segment monitoring, branching, and ALC ranking. Right: DeepLook on a higher-accuracy, lower-token-cost Pareto frontier, with an 87.3% average generated-token reduction relative to DeepConf-low (details in § 4.2 ).

Panel of Figure 2 — Figure 2: Uncertainty diagnostics correlate with errors on DeepSeek-R1-8B. Wrong traces (orange) exhibit more uncertain segments (a) and an earlier uncertainty onset (b) than correct traces (blue), motivating early confidence drops as triggers for lookahead exploration.

Figure 3: DeepLook pipeline. (1) Token-level top-

k

log-probabilities are aggregated into a segment confidence

\bar{C}(S_{i})

, which (2) is compared against a percentile threshold

\tau_{i}

over a local history window

\mathcal{H}_{i}

, so exploration fires only on a drop relative to the trace’s own recent history. (3) At a trigger,

K

branches are rolled forward over a horizon

H

and scored by Average Lookahead Confidence (ALC). (4) Branches are ranked and adaptively pruned, the best branch

b^{*}

resuming as the primary trajectory while the other retained branches are completed. (5) All completed traces are majority-voted into the final answer.

Model	Dataset	Path@1	Cons@512		DeepConf-low		DeepLook @128 (Ours)
Model	Dataset	Path@1	Tok	Acc	Tok	Acc	Tok	$\Delta\%\downarrow$ (C / DC)	Acc	$\Delta$ (C / DC)
DeepSeek-R1-8B	AIME24	83.0%	3.55	86.7%	0.78	92.5%	0.121	(-96.6%/-84.5%)	93.3%	(+6.6/+0.8)
	AIME25	76.9%	4.01	82.3%	1.24	86.4%	0.127	(-96.8%/-89.8%)	86.7%	(+4.4/+0.3)
	BRUMO25	80.0%	3.56	93.3%	1.07	90.0%	0.124	(-96.5%/-88.4%)	90.0%	(-3.3/+0.0)
	HMMT25	58.1%	4.49	69.8%	1.60	77.6%	0.213	(-95.3%/-86.7%)	73.3%	(+3.5/-4.3)
Qwen3-32B	AIME24	80.6%	2.00	84.8%	0.66	89.5%	0.102	(-94.9%/-84.5%)	90.0%	(+5.2/+0.5)

Table 1: Main results. Accuracy (%) and total generated-token cost (Tok,

\times 10^{8}

, all inference stages). For DeepLook @128,

\Delta\%\downarrow

and

\Delta

give token and accuracy change vs. (C / DC)

=

(Cons@512 / DeepConf-low). Shading marks the best and second-best accuracy per row. All accuracies are measured on

N{=}30

problems per benchmark.

Panel of Figure 4 — Figure 4: Illustrative case summary (not a verbatim trace). Hand-condensed decisive steps of an HMMT25 problem, contrasting the consensus failure mode of Cons@512 with the DeepLook run that recovered the correct answer; the summarization convention is described in Appendix F .

Figure 5: Average

n_{\text{unc}}

for correct vs. incorrect traces , DeepSeek-R1-8B on HMMT25, across four component variants under both readout regimes; G0 (greedy decoding) is shown as the two reference lines. The correct–incorrect gap ranges over

7.46

–

9.92

segments and never closes.

Panel of Figure 6 — Figure 6: Ablation and scaling analysis. (a) Accuracy as components are integrated (variants G0–G4 of § 4.3 ); $\Delta$ is the gain over Greedy. (b) Accuracy as a function of the lookahead horizon $H$ .

Appendix figures & tables18 assets

Supplementary material from the paper’s appendix.

Appendix

Hyperparameter	Value
Main segment length $L_{\mathrm{main}}$	512
Confidence top- $k$	20
Trigger window $W$	8 segments
Low quantile $q$	0.10
Warmup	4 segments
Hysteresis $\delta$	0.02

Appendix

Table 2: Default DeepLook hyperparameters (corresponding to DeepLook@32 ). The @64 and @128 variants increase only

K

(to 32 and 64); all other values are unchanged.

Panel of Figure 7 — Figure 7: Prompt logic for model_type=deepseek/qwen .

Panel of Figure 8 — Figure 8: Prompt logic for GPT-OSS models with reasoning_effort .

Asset type	Asset	Source/version used	License or terms
Dataset	AIME24 ( Maxwell-Jia, 2024 )	https://huggingface.co/datasets/Maxwell-Jia/AIME_2024 , 2024 dataset card	MIT License
Dataset	AIME25 ( MathArena, 2025a )	https://huggingface.co/datasets/MathArena/aime_2025 , 2025 MathArena dataset card	CC BY-NC-SA 4.0
Dataset	BRUMO25 ( MathArena, 2025b )	https://huggingface.co/datasets/MathArena/brumo_2025 , 2025 MathArena dataset card	CC BY-NC-SA 4.0
Dataset	HMMT25 ( MathArena, 2025c )	https://huggingface.co/datasets/MathArena/hmmt_feb_2025 , 2025 MathArena dataset card	CC BY-NC-SA 4.0
Model	DeepSeek-R1-8B ( Guo et al., 2025 )	https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B	MIT License
Model	Qwen3-32B ( Team, 2025 )	https://huggingface.co/Qwen/Qwen3-32B	Apache License 2.0

Appendix

Table 3: Licenses and terms for existing assets. Dataset and model licenses are taken from the linked public asset pages.

Figure 9: Accuracy vs. token cost trade-offs across benchmarks. Each panel plots dataset-level token cost against accuracy for one benchmark. Colors identify the base model, with DeepLook @128 shown as prominent stars and the corresponding Path@1 , DeepConf-low , and Cons@512 references shown with lighter markers. Faint colored segments connect Path@1 to DeepConf-low within the same model, visualizing the baseline accuracy–cost frontier that DeepLook @128 pushes beyond. Across AIME24, AIME25, BRUMO25, and HMMT25, DeepLook @128 consistently lies in the low-cost, high-accuracy region: it uses far fewer tokens than DeepConf-low and Cons@512 , while matching or exceeding their accuracy in most model–dataset settings. Path@1 is shown as the no-search reference point, plotted at the mean per-trajectory cost implied by Cons@512 (its token total divided by 512), which is the cost matching a single sampled trajectory.

Model	Dataset	Cons@512		DeepConf-high		DeepConf-low		DeepLook @32		DeepLook @64		DeepLook @128
Model	Dataset	Tok	Acc	Tok	Acc	Tok	Acc	Tok	Acc	Tok	Acc	Tok	Acc
DeepSeek-R1-8B	AIME24	3.55	86.7%	1.45	86.7%	0.78	92.5%	0.036	86.7%	0.064	86.7%	0.121	93.3%
	AIME25	4.01	82.3%	2.37	81.4%	1.24	86.4%	0.027	83.3%	0.072	83.3%	0.127	86.7%
	BRUMO25	3.56	93.3%	2.17	93.3%	1.07	90.0%	0.033	86.7%	0.065	90.0%	0.124	90.0%
	HMMT25	4.49	69.8%	3.43	70.0%	1.60	77.6%	0.040	66.7%	0.082	70.0%	0.213	73.3%
Qwen3-32B	AIME24	2.00	84.8%	0.88	86.4%	0.66	89.5%	0.018	86.7%	0.046	86.7%	0.102	90.0%

Appendix

Table 4: Full benchmark results. Accuracy (%) and token cost (Tok,

\times 10^{8}

) for Cons@512 , DeepConf baselines, and DeepLook budgets.

	Path@1	Cons@512		DeepConf-low		DeepLook @128
Model	Acc	Tok	Acc	Tok	Acc	Tok	Acc
DeepSeek-R1-8B	62.8%	10.9	66.2%	3.46	71.7%	0.198	64.6%
Qwen3-32B	68.9%	7.44	72.2%	3.21	73.0%	0.376	72.7%

Appendix

Table 5: GPQA-Diamond results (§ 4.6 ). Accuracy (%) and token cost (Tok,

\times 10^{8}

Component	% of Total
Main-path decoding	13.25%
Branch rollouts	23.65%
Lookahead evaluation	23.58%
Retained-branch completions	39.52%
Total	100.00%

Appendix

Table 6: Token cost breakdown on AIME25 (DeepSeek-R1-8B). Stage-wise share of DeepLook ’s dataset-level generated-token budget. The lookahead evaluation (ALC scoring) accounts for under 24% of total tokens.

Figure 10: Token- vs. segment-level confidence. Each segment spans two lines for readability (segment length $32$ tokens for illustration). The segment background encodes segment-level confidence $\bar{C}$ , where red backgrounds indicate uncertain segments. Individual token colors ( $C_{t}$ ) reveal local confidence fluctuations, motivating segment-level aggregation for robust detection of reasoning failures.

Panel of Figure 11 — Figure 11: Motivating analysis on Qwen3-32B. Incorrect traces show both more frequent uncertainty and earlier uncertainty onset than correct traces.

	DeepLook-Recursive		DeepLook
Dataset	Tok	Acc	Tok	Acc
AIME24	0.019	83.3	0.018	86.7
AIME25	0.039	76.7	0.038	80.0

Appendix

Table 7: Search strategy: DeepLook-Recursive vs. DeepLook . Tok (

\times 10^{8}

) and Acc (%) on Qwen3-32B.

Hyperparameter	Setting	Acc (%)	Tok (M)	Trig.
Ablation baseline		83.3	5.0	5.5
Robust: accuracy within one problem of baseline
Low quantile $q$	0.05	80.0	4.5	4.2
Low quantile $q$	0.25	80.0	7.5	7.3
Window size $W$	4	80.0	5.5	6.4
Window size $W$	32	80.0	6.9	8.5

Appendix

Table 8: Trigger parameter sensitivity on AIME25 (DeepSeek-R1-8B). One parameter varied at a time; all others held at the ablation baseline (

q{=}0.10

W{=}8

\delta{=}0.02

k{=}10

L_{\mathrm{main}}{=}512

, warm-up

4

K{=}16

R_{\max}{=}10

). This ablation baseline differs from the default configuration of Table 2 in

k

and

R_{\max}

, so its absolute token counts are not comparable to those of Table 9 . Tok in millions; “Trig.” is the mean number of trigger firings per problem.

Rounds	Tok ( $\times 10^{8}$ )	Acc (%)
2	0.0270	83.3
4	0.0437	83.3
6	0.0545	83.3
8	0.0649	83.3

Appendix

Table 9: Branching-round sensitivity on AIME25 (DeepSeek-R1-8B).

K=16

. Bold marks the default

R_{\max}

$\rho$	Tok ( $\times 10^{8}$ )	Acc (%)
0.10	0.0180	80.0
0.25	0.0438	80.0
0.50	0.0850	83.3
0.75	0.1204	83.3

Appendix

Table 10: Keep-ratio sensitivity on AIME25 (Qwen3-32B).

K=16

R_{\max}=2

; static keep ratio

\rho

Panel of Figure 12 — Figure 12: Illustrative summary: DeepLook exposes global constraints that self-consistency misses. In this HMMT combinatorics example, Cons@512 repeatedly selects the locally plausible odd-vertex heuristic and converges to 202. By intervening at the unstable pairing step, DeepLook uses lookahead to test whether the proposed edge removals remain globally consistent, detects the boundary-parity obstruction, and redirects the solution to the correct answer 200.

Panel of Figure 13 — Figure 13: Illustrative summary: DeepLook concentrates compute on fragile enumeration steps. In this AIME 2024 counting example, Cons@512 preserves a mechanically plausible inclusion-exclusion error and overcounts the rotational cases, giving 373. DeepLook identifies the low-confidence overlap calculation as the decision point, uses lookahead rollouts to reveal duplicated symmetry classes, and prunes the overcounted branch to recover the correct answer 371.

Panel of Figure 14 — Figure 14: Illustrative summary: the knowledge-boundary failure mode of § 4.6 . A hallucinated physical fact yields a coherent continuation that scores high ALC; internal confidence localizes instability, not falsehood.

Panel of Figure 15 — Figure 15: Illustrative summary: DeepLook rejects deceptive algebraic branches through lookahead verification. In this continuous-math example, Cons@512 treats the result of repeated squaring as reliable and outputs an extraneous root. DeepLook instead evaluates the downstream stability of that branch, checks the candidate against the original equation and domain behavior, and switches to the globally valid conclusion that no real solution exists.

Efficient Language Model Reasoning

Efficient language model reasoning studies how to improve problem-solving accuracy while reducing inference, memory, latency, and training costs.

See all 371 papers

Thinking by Subtraction: Confidence-Driven Contrastive Decoding for LLM Reasoning
Feb 20, 2026Lexiang Tang, Weihao Gao, Bingchen Zhao +4Efficient Language Model Reasoning LLM Reliability
Peking University, Beijing, China · University of Edinburgh, Edinburgh, UK.
Probability is Not Enough: Exploring and Counting Divergent Tokens for Reasoning Uncertainty Quantification in LLMs
Sep 29, 2026Feiyang Li, Shengjing Liu, Qi Zhan +7Uncertainty Quantification Confidence Estimation in Language Models
College of Computer Science and Software Engineering, Shenzhen University · Tsinghua University · Behavioral and Spatial AI Lab, Peking University & Tongji University
Think Deep, Not Just Long: Measuring LLM Reasoning Effort via Deep-Thinking Tokens
Feb 13, 2026Wei-Lin Chen, Liqian Peng, Tian Tan +5LLM Inference Efficiency Inference-Time Scaling
University of Virginia · Google

Feb 20, 2026J/K move · Enter open · S save

Thinking by Subtraction: Confidence-Driven Contrastive Decoding for LLM Reasoning

Lexiang Tang, Weihao Gao, Bingchen Zhao, Lu Ma+3

Peking University, Beijing, China · University of Edinburgh, Edinburgh, UK.

Efficient Language Model Reasoning LLM Reliability Contrastive Decoding Numerical Reasoning in Language Models

Open paper AskPDF arXiv

Recent work on test-time scaling for large language model (LLM) reasoning typically assumes that allocating more inference-time computation uniformly improves correctness. However, prior studies show that reasoning uncertainty is highly localized: a small subset of low-confidence tokens disproportionately contributes to reasoning errors and unnecessary output expansion. Motivated by this observation, we propose Thinking by Subtraction, a confidence-driven contrastive decoding approach that improves reasoning reliability through targeted token-level intervention. Our method, Confidence-Driven Contrastive Decoding, detects low-confidence tokens during decoding and intervenes selectively at these positions. It constructs a contrastive reference by replacing high-confidence tokens with minimal placeholders, and refines predictions by subtracting this reference distribution at low-confidence locations. Experiments show that CCD significantly improves accuracy across mathematical reasoning benchmarks while substantially reducing output length, with minimal KV-cache overhead. As a training-free method, CCD enhances reasoning reliability through targeted low-confidence intervention without computational redundancy. Our code will be made available at: https://github.com/bolo-web/CCD.

Abstract

Efficient Language Model Reasoning

Explore similar work

Thinking by Subtraction: Confidence-Driven Contrastive Decoding for LLM Reasoning

Probability is Not Enough: Exploring and Counting Divergent Tokens for Reasoning Uncertainty Quantification in LLMs

Think Deep, Not Just Long: Measuring LLM Reasoning Effort via Deep-Thinking Tokens