Inference-Time Compute Allocation

Follow this topic Ask about this topic

Momentum

10 papers in the last four weeks, against 2 the four weeks before. 0.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 53

Oct 6, 2026cs.CL

Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation

Prefill-only decision models inspired by the Jev model score every candidate in a menu during a single forward pass and never decode, which makes one call one to two orders of magnitude cheaper than a same-scale generative language model. We show that this read-out structure comes with a testable property. When an intervention changes only the candidate menu and leaves the input text fixed, the post-intervention accuracy is already determined by the cached first-pass distribution. The estimator restricts the pass-1 probabilities to the menu, renormalizes, and reads off the argmax; it uses no labels and no second forward pass. Across seven model families, ten datasets and two task types, menu-only interventions are predicted to within 4.2 points, and for one family the prediction is exact. A probability-level variant of the same estimator errs by 21.0 points, so the property lives in the ranking rather than in the probabilities and is not recovered by calibration. Same-scale generative language models do not share the property. On those models the same estimator errs by 1.6 to 15.8 points and degrades as the model grows. The property turns inference-time compute into a decision that can be made before deployment. Uniform extra passes buy calibration but almost no accuracy; at matched cost a confidence cascade outperforms every scheme that re-asks the same model, and curating the menu beats enlarging the model, with a 0.8B model on a curated 5-candidate menu reaching 95.4% on CLINC150 against 80.0% for a 4B model on the full 150-label menu.Code and data are available at https://github.com/rlisml/jev-cascade.

Ran Li, Lei Chen

Hong Kong University of Science and Technology · Hong Kong University of Science and Technology(Guangzhou)

Inference-Time Compute Allocation Efficient Language Model Inference

Oct 4, 2026cs.LG

Expanding LLM Reasoning

Extra inference compute is usually spent on sampling more reasoning chains. We study where inside an existing chain an additional continuation should begin. We define expansion utility, the change in correctness from restarting a chain at a stored step, and measure it at every eligible step for nine models on six benchmarks (41 model and benchmark cells). Restart position matters: steps selected on one set of continuations beat uniform placement when scored on disjoint ones, in held-out audits on 5, 16, and 38 cells (+4.25 points [+2.51, +6.63] in a fresh five-cell audit). A fixed rule that restarts from the last eligible steps, always-last, is a strong baseline: our learned router beats uniform placement but shows no detected gain over it, and on DeepSeek-R1-Distill-Qwen-14B/MATH-500 always-last exceeds the exact self-consistency frontier at matched aggregate generated output by +0.052 [+0.008, +0.098], using 0.774x the aggregate generated output of four-sample self-consistency. Cross-fitted oracle selection still finds held-out headroom beyond declared positional classes, a target for future selectors. Finally, breaking step-label ties by earliest index flips the sign of a pointwise selector's gain over uniform placement in every seed of a five-seed diagnostic with four rollouts per step; randomized ties remove the bias.

Rian Atri, Evan Luo

Keiji AI · University of California, Berkeley

LLM Inference Efficiency Inference-Time Scaling Efficient Language Model Reasoning Inference-Time Compute Allocation LLM Reasoning

Oct 1, 2026cs.AI

ReSolve: Reusing Candidate Reasoning through Selective Generative Moderation

Sampling multiple solutions spends computation on intermediate deductions and unfinished arguments as well as final answers. We introduce ReSolve, a training-free inference procedure that reuses this candidate reasoning through selective generative moderation. An answer-distribution controller invokes a model to examine existing derivations when candidates disagree or lack a parseable answer, then incorporates the generated solution into a bounded loop. Under Hybrid scoring on 130 competition-mathematics problems evaluated with two independently sampled candidate pools, ReSolve obtains 100 and 99 correct answers, compared with 91 and 92 for voting over the same four candidates, with no correct-to-incorrect changes relative to that vote in either pool. Eight-sample self-consistency obtains 94 and 96 correct answers while consuming substantially more tokens; ReSolve uses 46.3% and 47.2% fewer tokens in the two evaluations. A controlled ablation removes visible derivations while retaining answer keys, vote counts, and the per-state output-cap rule, reducing accuracy from 100 to 93 correct despite increasing computation. Selective and always-on Uniform moderation both solve 97 problems, while selectivity reduces moderation tokens by approximately 54% and total pipeline tokens by 6.2%. These results support candidate reasoning as reusable inference computation. They do not establish an accuracy advantage over additional sampling or a distinct benefit from specialized route instructions.

Bangji Yang, Jiajun Fan, Hongba Ma +6

University of Illinois at Urbana-Champaign · Tsinghua University · Rutgers University +2

Efficient Language Model Reasoning Inference-Time Compute Allocation Efficient Language Model Inference LLM Mathematical Reasoning

Sep 30, 2026cs.AI

Budget Boundary Effects in Test-Time Mathematical Reasoning

A cumulative token cap can fall inside a mathematical derivation, forcing a test-time controller to choose between stopping at the cap (strict) and allowing the current attempt to finish (advisory). We measure this boundary choice with paired offline replays of 19,200 public traces: 120 AIME, BrUMO and HMMT problems and two archive configurations of one model. Candidate order and a 16-attempt cap are fixed, and answer selection is blind to reference answers and correctness labels. Three findings emerge. First, at the 4k cap, most advisory accuracy gains replace abstention with a correct answer; strict stopping pays for an unfinished prefix that the completed-only selector cannot use. Second, comparisons along realized cost differ from same-cap comparisons: advisory 4k in low has higher accuracy than strict 8k at comparable mean completion cost, while in high its observed accuracy is 0.42 points below strict 32k using 59% of its mean tokens. These aggregate comparisons do not establish equal-compute superiority or accuracy equivalence. Third, increased candidate coverage does not guarantee higher answer accuracy: a log-probability selector loses accuracy while coverage rises, including after a source-grade consistency repair. Same-cap majority-accuracy differences shrink below 1.3 percentage points at 32k. Budget curves should jointly state the cap, realized cost, eligible candidates, stopping rule and selector information.

Guilin Zhang, Ziqi Tan, Wulan Guo +5

Yantong AI

Mathematical Reasoning Benchmarks Test-Time Optimization Optimal Stopping Inference-Time Compute Allocation LLM Mathematical Reasoning

Sep 29, 2026cs.AI

Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning

As agents take on longer and more complex problems, controlling the execution becomes a task in its own right. Each step in the run brings new control choices, like which partial work to build on, whether to start fresh, or when to stop. We introduce agentic meta-reasoning, an inference-time harness that makes these choices an explicit and structured reasoning process. Workers carry out the task-level computation, while a controller consolidates what the run has established, explores next options, assesses what each option is worth under the remaining budget, and dispatches the chosen work with context drawn from persistent memory. Between decisions the controller carries only a compact account of the run rather than replaying its full history. Our baselines span production coding agents and research harnesses, together with a Direct Control Agent using the same workers and compute budget allowance. On ProgramBench, which tests long-horizon agentic capability through program reconstruction, meta-reasoning achieves 71.5% with GPT-5.5 against 58.0% for Codex; with Opus 4.8 it achieves 67.2% against 65.5% for Claude Code. On the other benchmarks, spanning abstract reasoning, multi-domain long-horizon reasoning, and proof generation, it gains between 3.6 and 4.2 points over direct control, averaged across three frontier models. It keeps improving over the tested budget ranges where direct control plateaus, though its overhead can hurt at small budgets. Artifact-graph analysis reveals more reuse of earlier work, higher coverage of correct solutions in most settings, and nonuniform gains in final selection. These results indicate that spending computation on structured control becomes more important as agents scale to longer runs.

Paras Dahal, Anton Bakhtin, Taco Cohen +9

Meta Superintelligence Labs

Agentic Reasoning Inference-Time Compute Allocation Long-Horizon LLM Agents LLM Agent Harnesses

Sep 28, 2026cs.LG

The Decision Value of Perception Compute

Adaptive perception spends extra computation on inputs where perception is expected to improve. When perception feeds a downstream decision system, a better perception output need not produce a better decision. We define the decision value of perception compute as the change in downstream loss from escalating an input from a cheap to an expensive perception mode. Because this value can be negative, the allocation of perception compute should be judged against a budget-constrained decision oracle, with uniform full-fidelity inference as a baseline rather than an upper bound. We introduce DEEP (Decision Evaluation for Escalated Perception), a benchmark that scores pre-escalation allocators against this oracle under selection, latency and energy budgets, charging each allocator for its own computation. With deployed monocular geometry on KITTI and nuScenes, we find that 34--54% of the escalations that change downstream loss make it worse; harmful escalations also occur for the published PDM-Closed planner, evaluated open-loop on nuPlan with real detector outcomes. On nuScenes, perception-level gain frequently disagrees in sign with decision value. This mismatch has practical consequences: choosing among fixed deployable signals by missed-object perception gain rather than by decision value reduces realized test decision gain by 7.4% of the all-cheap loss on average. Learned allocators recover part of the oracle's value by finding beneficial escalations but select nearly as much harm as random, and once their own computation is charged at a 20% latency budget, only the lightweight routers, at about 3.5% of a full detector pass, still beat random.

Hoang Pham Cong, Ho Viet Duc Luong

Hanoi University of Science and Technology (HUST), Hanoi, Vietnam

Cost-Aware Inference Autonomous Driving Perception Inference-Time Compute Allocation

Sep 27, 2026cs.RO

ReSync: Re-Aligning the Two Clocks of Asynchronous World-Action Models

Jointly generating future video and actions has become a standard recipe for world-action models, and the strongest systems denoise the two streams on separate schedules: actions are decoded in few steps so control stays fast, while the video stream runs longer to keep the predicted future sharp. The design is deliberate, but it leaves the two streams on different clocks, and an action can become executable while the future that should justify it is still largely unresolved. We formalize this as a two-clock view of asynchronous inference and introduce the commitment-evidence gap, a quantity read directly from a model's own sampling schedule rather than measured by search. The gap is predictive: as it widens, candidate utility becomes harder to identify and extra candidate sampling buys less, while advancing the world stream buys more, and the two cross. Spending more world computation is therefore not simply better. The useful interval is closed at both ends, and both ends can be read off the schedule before any rollout. ReSync places the computation inside it: hold the action state, advance only the world within the supported window, then resume native denoising. No parameters change and no candidates are compared. On a frozen paired RoboCasa panel this improves success by 4.48 points, while an equal-compute control that waits without advancing the world does not move, and the same rule transfers to a second benchmark and a second backbone without retuning.

Xi Lin, Feihong Zhang, Yulong Shi +6

Johns Hopkins University · Tsinghua University · Yinwang Intelligent Technology Co., Ltd. +1

Action-Conditioned Video Generation Video Diffusion Models Action-Conditioned World Models Inference-Time Compute Allocation

Sep 17, 2026cs.AI

The Organization of Inference: Information, Resource Constraints, and AI Production

The economic value of inference depends on how capacity and task information are distributed across stages of AI production. We study these organizational margins using controlled workflow experiments on externally verified software-engineering tasks. In two matched resource panels, direct execution records the same success rate of 59.6 percent at logical-token ceilings of 12,000 and 24,000, while success under information-constrained planning rises from 36.2 to 51.2 percent. The planning disadvantage narrows by 15.0 percentage points (95 percent task-cluster bootstrap interval: 4.2 to 25.8). A strict read-only planning campaign varies whether the planner sees the task issue. At 12,000 tokens, issue access raises success by about 16 percentage points over issue-hidden planning. Compared with direct execution, task-informed planning is about 10 points lower at 12,000 tokens; at 24,000 tokens, it shows a 29.6-point advantage. In the resource panels, direct execution uses substantially less than either ceiling, while the planning workflow's binding rate falls from 46.2 to 0.8 percent and downstream execution accounts for 89.9 percent of the increase in total use. Scale determines the capacity available to a system; workflow and information structure shape the productive value

Yukun Zhang, Kemu Xu, Yishen Chen

The Chinese University of Hong Kong Hong Kong, China · University of Edinburgh Edinburgh, United Kingdom · The Chinese University of Hong Kong, Shenzhen Shenzhen, China

LLM Planning Inference-Time Scaling Inference-Time Compute Allocation

Sep 14, 2026cs.CL

When Agents Slow Down: Understanding LLM Agents' Test-Time Strategies via Elo-per-token Analysis

Large language model (LLM) agents allocate test-time compute adaptively as they revise solutions, use tools, explore alternatives, and decide when to stop. This test-time strategy makes it difficult to measure how agent performance scales. We study open-ended tasks that provide continuous scores for intermediate submissions, making progress observable throughout long trajectories. We propose Elo-per-token analysis, which tracks the best solution found at each token budget and uses a Bradley-Terry model to aggregate within-task orderings into Elo ratings across tasks with different score scales. We apply it to four general-purpose agents on four open-ended benchmarks, with sessions of up to 100M tokens, and to three feedback-driven LLM optimization harnesses in controlled single-task interventions. Independent sampling provides a theoretically characterized reference, for which Elo grows linearly with log compute. Against this reference, agents can initially convert tokens into Elo faster than independent sampling, but their marginal gains diminish and eventually fall below the reference. In contrast, the strongest historical human contestants improve superlinearly over contest time on shared AtCoder Heuristic Contest tasks, providing evidence of continual learning and substantial headroom after agents slow down. We define the scaling inflection point as the per-session budget where marginal Elo gains match the independent-sampling reference. Using this point as the per-session budget, we split 100M tokens across parallel sessions on FrontierCS Polyomino Packing, gaining +264 Elo over one long session and +355 over ten short sessions.

Kaiyuan Liu, Qiuyang Mang, Bo Peng +6

UC Berkeley · University of Washington · Princeton University +1

LLM Agent Evaluation Test-Time Scaling Long-Horizon Agent Evaluation AI Agent Benchmarks Inference-Time Compute Allocation

Sep 14, 2026cs.CL

Intelligence Under Time Constraints: Rethinking Test-Time Compute

Intelligence under time constraints requires deciding not only how much to compute, but when computation is worth starting. We study this problem in streaming interactions, where evidence arrives incrementally and may be revised. Early computation has more time to finish but rests on incomplete evidence; waiting improves information while shrinking computational slack. We call this the information-slack dilemma. We take the evidence-dependent computational job as the unit of analysis: when to start it, what supports its result, and when that result can be committed. Advance computation is valuable only insofar as its benefits survive the costs of verification, invalidation, and recovery. This applies to grounded incremental processing and reusable preparation as well as future-dependent speculation. We propose a research agenda on computation under evolving evidence, prioritizing selective recovery under controlled evidence revisions. Evaluation should separate earlier-execution effects, deployment value against a full-input alternative, and the added value of predictive policies, while accounting for shared-resource costs. The objective is not maximal advance computation, but more trustworthy, on-time responses within a declared resource envelope.

Xiaotian Zhang

Trooly.AI

Inference-Time Scaling Inference-Time Compute Allocation Online Resource Allocation

Sep 12, 2026cs.AI

Routing by Reasoning Need: Trajectory-Aware Decoding Control for Diffusion Vision-Language Models

Diffusion vision-language models generate answers through iterative refinement, exposing intermediate answer trajectories that can be inspected and controlled at inference time. However, this controllability creates a reasoning-need mismatch, where a universal generation length is applied to questions with different reasoning demands. Visually closed questions may be harmed by continued refinement after a stable answer has formed, whereas reasoning-sensitive questions may be harmed by premature commitment. We formulate this problem as reasoning-budget mismatch and study it in LLaDA-V. Rather than choosing a universal generation length, our training-free controller routes each example to early commitment, baseline preservation, or reasoning-supportive decoding using trajectory signals from answer closure, commitment evidence, and representation revision pressure, without using ground-truth answers. Across answer-focused, mixed-reasoning, and CoT-sensitive benchmarks, routed control improves robustness over fixed long decoding, pure short decoding, and single-rule interventions. The gains are not explained by shorter outputs alone. Answer-closed examples often benefit from commitment, whereas CoT-sensitive examples require preserving or supporting intermediate reasoning. Taken together, these results suggest diffusion VLM decoding should route inference-time control by the state suggested by the observed trajectory instead of relying on a universal decoding length.

Yixiang Liu, Zhongxing Xu, Zhonghua Wang +1

Southern University of Science and Technology · Monash University

Diffusion Language Model Inference VLM Reasoning Diffusion Language Model Decoding Inference-Time Compute Allocation Diffusion Language Models

Aug 13, 2026cs.GT

Keep, Customize, or Exit: Default Design and Token Pricing in LLM Reasoning Services

We study a large language model (LLM) service in which a provider chooses a per-token price and a default reasoning-token allocation, while a user may accept the default, customize the allocation, or exit. Larger allocations can improve accuracy but increase token cost and latency. We model this interaction as a Stackelberg game and derive the user's unique optimal customized allocation in closed form. For any price, the acceptable defaults form either an empty set or a compact interval. We characterize the provider's optimal default through a three-regime rule, reduce equilibrium computation to a one-dimensional price optimization, and prove the existence of the equilibrium. We further show that defaults affect the implemented reasoning allocation only when users value the convenience of avoiding customization; otherwise, every service-providing outcome implements the user's optimal customized allocation. Experiments with two compact open-weight reasoning models on five mathematics and science benchmarks support the accuracy-token model and show how model and task characteristics determine equilibrium prices, defaults, and reasoning allocations.

Ahmet Bugra Gundogan, Yigit Turkmen, Melih Bastopcu

Department of Electrical and Electronics Engineering

LLM Serving Inference-Time Compute Allocation

Aug 13, 2026cs.LG

Sampling Luck Masquerades as Allocation Gain: Auditing Test-Time Budget Allocation for Neural Combinatorial Optimization

Neural combinatorial optimization (NCO) solvers report the best of many sampled solutions per instance, and the sample count is, by convention, identical for every instance. Whether a non-uniform allocation of a fixed total budget would buy anything has not been measured. We measure it, and we audit the measurement itself. First, on in-distribution workloads the allocation headroom is not detectable. Across three pretrained solvers (POMO, AM, SymNCO) on uniform TSP-100, an oracle allocation computed and evaluated on the same stored samples reports a 2.2-2.6% gain with intervals excluding zero; measured out of sample the same gain is indistinguishable from zero (0.457, 0.015, -0.512 percent). Following the customary in-sample procedure, all three solvers would have supported a published 2%-level gain that does not exist. We calibrate this bias against an instance-wise null in which the true gain is zero by construction; over the ranges we test it does not shrink with more samples or more instances. Second, the same correction that removes the phantom gains preserves a real one. Under distribution shift (a workload mixing uniform and clustered instances), a pre-registered confirmatory experiment finds that allocation guided by held-out sample statistics improves best-of-k by 11.5% (AM, primary endpoint; 95% CI [7.4, 19.7]) and 12.0% (SymNCO, replication) at equal evaluation budget, with the signal-acquisition cost not charged; a pre-registered negative control (POMO, an order of magnitude more robust to shift) shows -0.3% [-0.7, 0.24]. The gain exceeds a frozen distribution-label baseline by 4.2 points [1.9, 7.7]. An exploratory policy charging a 20-sample probe against the same budget retains 3.4% (AM) and 4.6% (SymNCO). We give a correction procedure and a reporting checklist, and release all data, code, and the pre-registration record.

Jinhyung Bae

Department of Industrial and Management Engineering and the AI Data Convergence program, Hankuk University of Foreign Studies, Yongin, Republic of Korea

Algorithmic Auditing Inference-Time Compute Allocation Selection Bias Neural Combinatorial Optimization

Aug 8, 2026cs.AI

Thought-Level Beam Search for Reasoning

Test-time compute scaling is a primary driver of performance in large reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from \emph{how much} compute to spend, to \emph{where} to allocate it. We formalize test-time reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to actively allocate the compute to the most promising partial progress: traditional parallel sampling treats traces independently and induces severe memory bottlenecks, while subtractive pruning starves hardware and fails to actively and sufficiently shift the output distribution. To overcome this dichotomy, we introduce Gambit, an inference algorithm that executes \emph{thought-level beam search}. By periodically pruning unpromising trajectories and immediately branching from high-quality prefixes, Gambit dynamically concentrates compute onto the most promising reasoning traces via a light-weight scorer probing hidden states while maintaining continuous high hardware utilization. Extensive evaluations across multiple models and benchmarks demonstrate that Gambit strictly dominates existing baselines. Under identical hardware constraints, our method yields up to a +6.7% absolute accuracy gain on HMMT-24 and +3.3% on AIME-25 over pruning baselines, delivers

>2\times

higher throughput on trace completion, and reduces total token consumption by up to 68.5% relative to standard parallel sampling.

Lijie Yang, Hongyin Luo, Jiawei Zhao +2

Princeton University · MIT CSAIL · Meta AI

Inference-Time Search Efficient Language Model Reasoning Inference-Time Compute Allocation Efficient Language Model Inference Algorithmic Reasoning Large Reasoning Models

Aug 8, 2026cs.CL

Thinking Hard, Not Smart: Reasoning Models Fail to Ration Test-Time Compute Across Questions

Reasoning language models increasingly use test-time compute to improve performance, but existing evaluations typically study this compute one question at a time. Yet when multiple problems share an end-to-end cost or latency constraint, models must decide how to divide limited inference compute among them. We introduce an exam-style evaluation framework for studying this setting, in which a model must distribute one shared token budget across questions with different difficulty and point values to maximize its total score. Across several open and frontier reasoning models, we find that models fail to allocate a shared budget strategically across questions of varying difficulties and values. Models behave largely as greedy sequential solvers: they prioritize questions by presentation order, front-load effort on early questions, and remain insensitive to value, with these tendencies becoming more pronounced as the number of questions grows. Explicit planning prompts spread compute more evenly but do not produce value- or difficulty-aware prioritization. The same behavioral pattern extends from mathematical to code reasoning. These findings establish global budget allocation as a distinct capability that is not captured by conventional per-question evaluation and remains a challenge for current reasoning models.

Chenrui Fan, Yize Cheng, Ming Li +3

University of Maryland, College Park · MBZUAI, UAE

LLM Evaluation Inference-Time Compute Allocation Large Reasoning Models

Aug 7, 2026cs.AI

CoBa: Cost-Effective Test-Time Scaling via Compute-Balanced Routing

Test-time scaling is often implemented by spending more compute along one axis: sampling more solutions, extending a chain of thought, or applying a stronger evaluator. Under a fixed inference budget, these choices compete. This paper formulates test-time reasoning as a compute-allocation problem in which a system must decide whether the next unit of compute should be spent on generation, verification, or stopping. We introduce CoBa, a compute-balanced routing policy that first obtains a small set of candidates, applies cheap verification broadly, and routes uncertain or high-value candidates to stronger verification. On 3,129 example-generator evaluations spanning MATH-500, AIME 2024/2025, AMC 2023, and procedural symbolic reasoning, CoBa-Routed-Strong reaches 85.13% macro accuracy, statistically matching a self-evaluation weighted-voting proxy at 85.20% while using 49.1% fewer parameter-weighted tokens. It also matches best-of-16 majority voting within 0.01 macro-accuracy points while using 58.9% fewer parameter-weighted tokens; paired tests retain a small best-of-16 edge at substantially higher cost. Paired bootstrap tests show significant gains over single-sample decoding, while the remaining gap to the pool oracle exposes headroom for sharper routing. For local reasoning systems, test-time scaling becomes a question of where the next computation is most valuable.

Yan Zhou, Yue Ouyang, Kaiyang Zheng +1

School of Mathematics and Statistics, Changsha University of Science and Technology, Changsha, China · School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China.

Cost-Aware Inference Test-Time Scaling Inference-Time Compute Allocation Algorithmic Reasoning

Jul 30, 2026cs.AI

Rethinking Inference-Time Scaling in Local Computer-Use Agents: Failure Modes and Compute Tradeoffs

Deploying autonomous computer-use agents (CUAs) locally is increasingly important for privacy, cost efficiency, and practical usability, yet improving their performance under strict hardware constraints remains challenging. While recent studies show that inference-time scaling can improve frontier computer-use agents through additional computation during execution, its effectiveness for resource-constrained local models remains poorly understood. We present a systematic empirical study of inference-time scaling in local CUAs across contextual, temporal, structural, and parallel dimensions. We evaluate Qwen3-VL-8B/30B-A3B, UI-TARS-1.5-7B, and OpenCUA-7B on the OSWorld benchmark. Our results show that additional computation often yields diminishing returns while changing failure modes. Contextual scaling provides historical grounding that improves trajectory stability and task accuracy, but its gains saturate as token cost increases and failures shift from repetitive or stalled trajectories toward premature false successes. Temporal scaling similarly reduces max-step stalls, yet does not substantially improve task success, indicating that longer horizons often extend erroneous trajectories rather than correct them. We further find that structural decomposition can introduce planning and formatting overhead in local two-stage agents, while parallel scaling partially mitigates these failures at a substantial computational cost. Overall, our findings suggest that efficient local CUAs require selective compute allocation, failure-aware control mechanisms, and agentic frameworks designed around the capabilities and limitations of local models.

Woongkyu Lee, Jungwook Choi

Hanyang University

Computer-Use Agents Inference-Time Scaling Inference-Time Compute Allocation On-Device Inference

Jul 30, 2026cs.AI

Search as Computation Allocation

Many algorithms spend an internal resource before returning a decision and are evaluated only by the quality of that terminal output. We formalize such procedures as terminal computation-allocation problems: costly computations produce observations, update beliefs about a latent environment, and matter only through terminal decision loss. Bellman equations characterize optimal allocation under fixed budgets, priced computation, and exact certification. We then relate value of computation (VOC) to information. Mutual information equals myopic VOC under log loss, whereas under simple regret VOC is a knowledge-gradient quantity; moreover, information gain can rank computations arbitrarily poorly, although it gives a one-sided upper bound on VOC. Bandit pulls, tree simulations, and node expansions illustrate the same model under different computation topologies. Finally, under an explicit frontier-resolution and heuristic-error model, maximizing approximate VOC recovers weighted A*, with A* and greedy best-first search as limiting cases. The theory identifies a shared decision problem without asserting that one acquisition rule is universally optimal.

Alexander Tuisov

Technion - Israel Institute of Technology

Inference-Time Compute Allocation

Jul 27, 2026cs.AI

Exploring Budgeted Image Classification with Content-Sensitive Resource Allocation

The ever-growing adoption of Artificial Intelligence (AI) creates the need to deploy Deep Neural Networks in a variety of computational environments. We consider dynamic environments, where computational requirements are subject to change, and we pose the following question: How do we adjust the complexity of an AI classification system, in order to maximize its accuracy, while meeting changing computational constraints? We call this problem Budgeted Image Classification, and we formally formulate it as a resource allocation integer program. Given a computational budget, a batch of images, and a classification system that can make decisions with varying complexity (it has multiple decision points), we explore strategies to allocate images to decision points, in order to maximize accuracy within the available budget. The original integer program is NP-Hard, so, we propose a continuous relaxation, leading to a content-agnostic allocation strategy which assigns images to decision points without considering their particular content. We address this issue by proposing a content-sensitive strategy, that we experimentally show it leads to superior performance. We theoretically study the behavior of our strategies, deriving conditions that must be satisfied by decision points to be suitable for budgeted classification. We analyze fails cases, offering insights for future research directions.

Athanasios G. Papadopoulos

Tandon School of Engineering, New York University

Cost-Aware Inference Classification Inference-Time Compute Allocation Image Classification Adaptive Computation

Jul 9, 2026cs.LG

Resample or Reroute? Recoverable Stopping Debt Without Identified Action Selection

After a weak verifier accepts a large-language-model response, a second call may resample or reroute. Because correctness is hidden, action selection is an identification problem. We order three gates: recoverable stopping debt, two-sided FIT action support, and held-out value from an outcome-blind selector. In a pinned 152-query MBPP+ experiment, a Qwen2.5-14B Base-only false-positive stop leaves +2.592 percentage points of Qwen2.5-7B recovery (query-cluster 95% interval [+1.618, +3.664]). Separately, after 7B Base-test rejection, fixed escalation to 14B exceeds leave-one-out 7B resampling by +2.882 points [+0.931, +5.201]; this is fixed-action ranking, not conditional selection. An all-episode audit produces a +2.697-point realized-maximum gap, but for two actions this statistic equals (1/2)E|Delta| - (1/2)|E Delta| and contains no observable-history term. It lies inside an exact-fold exchangeable reference (mean +3.158; 95% interval [+2.434, +3.947]). The audit unconditionally acts on 1,520 episodes: 1,240 observable stops and 280 verifier rejections; 198 stops are evaluator-only false positives. Neither tested outcome-blind controller improves on fixed rerouting. A separate LiveCodeBench ladder has all-zero FIT action advantages despite exclusive TEST rescues. A preregistered BigCodeBench support gate then finds only 23/19 and 22/19 signed episodes/queries against minima of 25/20, so L1-L4, DEV, and TEST stay unopened. Stopping debt exists, but current evidence does not identify when to resample rather than reroute.

Teng-Ruei Chen

Institute of Bioinformatics and Systems Biology, National Yang Ming Chiao Tung University, Hsinchu, Taiwan · Krixvon, Taipei, Taiwan

Code Generation LLM Reliability Inference-Time Compute Allocation LLM Routing

Jun 30, 2026cs.RO

ELASTIC: Efficiently Learning to Adaptively Scale Test-Time Compute for Generative Control Policies

Generative control policies (GCPs), such as diffusion policies and flow-based vision-language-action models, enable test-time scaling in robot control. Test-time compute can be allocated along two axes: sequential scaling, which increases denoising steps to refine actions, and parallel scaling, which samples multiple candidate actions to search across modes of the policy distribution. However, the optimal allocation of sequential and parallel compute is hard to know a priori as it is state-, task-, and policy-dependent. For example, early stages of a grasp may benefit from broader parallel exploration, while near-contact phases may require more sequential refinement for precision. We present ELASTIC, an algorithm that learns state-dependent test-time compute schedules for GCPs. We formulate compute allocation as a meta-Markov Decision Process in which a meta-policy interacts with a frozen pretrained robot policy and selects sequential steps and parallel samples at each denoising iteration to maximize task success while minimizing compute. Using reinforcement learning, this meta-policy also learns adaptive compute schedules without access to the GCP's training data. Across simulated manipulation benchmarks with diffusion policies, ELASTIC Pareto-dominates fixed and single-axis scaling baselines at matched compute budgets. On real-world robot manipulation with the

π_{0.5}

vision-language-action model, ELASTIC matches best-of-

10

success while reducing wall-clock latency by 34%.

Andrew Zou Li, Gokul Swamy, Yonatan Bisk +1

Carnegie Mellon University

Diffusion Policy Test-Time RL Robotic Manipulation Inference-Time Compute Allocation

Jun 26, 2026cs.IR

End-to-End Dynamic Sparsity for Resource-Adaptive LLM Inference

Large Language Models (LLMs) inference is typically deployed under a static resource assumption, where models execute a fixed computational graph regardless of the runtime environment. However, real-world cloud infrastructure is inherently dynamic, characterized by fluctuating availability (e.g., spot instance preemption) and tiered Quality-of-Service requirements. In such volatile settings, static models are inflexible: they either crash under resource constraints or waste compute on redundant operations. To bridge this gap, we propose Learning to Allocate (L2A), an end-to-end framework for resource-adaptive inference. Unlike prior methods that condition only on input difficulty, we formulate inference as a constrained allocation problem conditioned on both the input and the runtime resource budget itself. We introduce lightweight, budget-conditioned and input-aware gating networks integrated into the LLM. These gates are trained via a unified objective that jointly optimizes task performance, logical consistency, and resource costs along three axes matching how real-world dynamics manifest: layer skipping for memory and depth pressure, head pruning for throughput contention, and reasoning-token reduction for latency tightening. This lets the model learn a budget-aware policy beyond input difficulty alone: it adaptively configures its computational footprint with respect to real-time resource dynamics, maximizing reasoning depth when resources permit while enforcing strict frugality when budgets tighten. A single L2A model traces the entire compute-accuracy Pareto frontier on Llama-3-8B and Qwen-3-4B: at up to 34% realized layer sparsity, it stays within 0.6% of the dense baseline on GSM8K, with the same gap holding zero-shot on out-of-distribution tasks, while every static or heuristic baseline requires a separately tuned model and still drops by 5-10% at comparable inference time.

Yuhang Chen, Jinhao Duan, Ruichen Zhang +11

University of North Carolina at Chapel Hill · Meta AI

LLM Inference Structured Sparsity LLM Inference Acceleration Inference-Time Compute Allocation Sparse Neural Networks Conditional Computation

Jun 25, 2026cs.AI

Humans Disengage, Reasoning Models Persist: Separating Difficulty Registration from Deliberation Allocation

Large reasoning models (LRMs) tend to produce longer reasoning traces for problems on which humans also spend more time. This correspondence suggests a shared sensitivity to difficulty, yet difficult problems can invite both persistence and withdrawal. We distinguish difficulty registration, expressed in which problems elicit more deliberation, from the allocation of further work. We examine their relation in matched human and LRM data from visual abstraction, intuitive physics, and relational reasoning. On visual abstraction, model trace length tracks the human ordering of problems by duration. After item identity is controlled, successful human attempts last longer than failed attempts, while failed LRM attempts have longer traces than successful ones in the pooled model analysis. The estimated outcome slopes follow the same pattern in intuitive physics. In relational reasoning, successful attempts are longer in separate human and model analyses. Fitting the two groups together with shared item effects yields a human-LRM difference in the relation between duration and outcome. Human grid actions connect longer attempts with sustained task engagement. Failed LRM traces contain more hedging or repetition after length is controlled, with the form of the difference varying across tasks. A resource-rational account explains how the expected reducibility of uncertainty and the value assigned to further computation can produce different patterns of persistence despite similar sensitivity to difficulty. Agreement about which problems require more deliberation can therefore coexist with different patterns of persistence on those problems.

Han-yu Wang

The University of Hong Kong

Computational Cognitive Modeling Inference-Time Compute Allocation LLM Reasoning Large Reasoning Models

Jun 22, 2026cs.CR

Maestro Order: A Model-Agnostic Orchestration Harness

A single forward pass of a capable model is a fast, fluent, and unreliable problem-solver: it is right often enough to be useful and wrong often enough to be dangerous; in language models, such confident errors are known as hallucinations. We present Maestro Order, a model-agnostic orchestration harness that turns unreliable solvers into reliable problem-solving systems by composing them according to four structural primitives (decompose, ensemble, verify, and recurse) and a budget-aware controller that decides where to spend compute. The harness treats any model as a black-box base solver behind a uniform interface, layers a verifier ensemble whose discrimination is measured online, and allocates verification and voting to the stages with the highest marginal reliability per unit cost. We give the architecture, the message and state schema, the controller algorithm, and the engineering that makes it deterministic, observable, and fault-tolerant. We then specify an evaluation methodology (reliability at fixed cost, coverage, calibration, and ablations) and report results from a faithful Monte Carlo simulation of the harness over a parameterized solver/verifier model. The simulation reproduces the predicted laws quantitatively: verification amplifies reliability geometrically (e.g.

0.55\to0.98

with two gates,

\to0.999

with four), voting helps only above chance and is limited by shared errors, and a budget-aware controller reaches a target reliability at a small fraction of the cost of voting alone by selecting the cheapest mechanism for each regime. We close with failure modes (verifier gaming, correlated errors, and decomposition error compounding) and concrete guidance: build robust checkers, diversify solvers, and let the controller put compute where the information is.

Hidayet Aksu

LLM Agent Orchestration LLM Reliability Inference-Time Compute Allocation Language Model Ensembles

Jun 14, 2026cs.AI

Heteroskedastic Signals in Budgeted LLM Verification: Structural Heterogeneity Limits Optimization Gains

Large language model (LLM) systems increasingly use uncertainty signals to allocate limited computation across verification, test-time scaling, tool execution, and other selective-compute decisions. Such policies rely on a \emph{global signal comparability assumption}: equal scores should carry comparable decision value across inputs. Using budgeted verification as a controlled diagnostic setting, we identify a failure mode of this assumption: uncertainty quality is heteroskedastic across cost strata, with some regions exhibiting near-random discriminability despite concentrating many errors. Under an explicit local model, we characterize the resulting distortion of global allocation and show that its upper bound scales with cross-stratum signal-quality dispersion. We separate weak signals, optimization instability, and structural heterogeneity through a controlled intervention hierarchy: Threshold, MP-Adapt, MP-Strat, and a deliberately simple cost-stratified thresholding intervention (CST). Across MBPP and MATH using Qwen3-8B, LLaMA3-8B, and GPT-4o-mini, global online adaptation yields inconsistent gains over static thresholding; MP-Strat partially recovers performance, while CST improves hit rate by up to 17 percentage points in strongly heterogeneous settings without gradient updates. These results identify structural heterogeneity, rather than optimizer weakness alone, as the primary bottleneck in the observed settings. More broadly, misaligned feedback structure cannot always be repaired by stronger optimization.

Jinlong Yang

Northwestern Polytechnical University

Large Language Model-Guided Optimization LLM Uncertainty Estimation Inference-Time Compute Allocation

Jun 10, 2026cs.RO

DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners?

Vision-Language Models (VLMs) are increasingly deployed as high-level planners for embodied agents, with an emerging strategy of scaling test-time compute to improve capability. However, we observe that doing so increases latency, token usage, and FLOPs while yielding uneven, often diminishing gains in downstream success, limiting where embodied agents can be deployed. We argue that choosing when and where to spend test-time compute is central to bringing frontier performance to the real world. We introduce DIRECT, a routing framework that uses multimodal scene context to allocate compute per prompt, improving the success--cost Pareto frontier over fixed model selection. Across three dominant scaling axes, namely chain-of-thought depth, model size, and memory history, our experiments on VLABench and RoboMME show that test-time compute is not a uniform lever: different axes yield qualitatively distinct capability gains. We validate these insights on a physical Franka arm in a DROID setup spanning zero-shot manipulation and long-horizon chaining, where our router matches or exceeds a stronger model's success rate at up to 65% lower average latency. Ultimately, our results show that naively scaling test-time compute is wasteful, and that DIRECT can provide frontier-level embodied planning in robotic systems at a fraction of the cost. Project page can be found at jadee-dao.github.io/direct/.

Jadelynn Dao, Milan Ganai, Yasmina Abukhadra +7

Stanford University · University of Waterloo · NVIDIA

Test-Time Scaling for VLMs Inference-Time Scaling Inference-Time Compute Allocation VLMs for Robotics

Jun 7, 2026cs.LG

Intrinsic Selection and Particle Resampling for Inference-Time Scaling Beyond Domain Verifiability

Inference-Time Scaling (ITS) has largely succeeded in verifiable domains like math and coding, where cheap verification enables scalable output selection. However, extending ITS to tasks prone to systematic failure - driven by faulty initial assumptions or unmet multidimensional constraints - typically relies on costly external solvers or brittle, model-based verifiers. Our key insight is that the intrinsic statistics of parallel sample sets, specifically length-adjusted tail entropy, provide a robust discriminative signal for solution quality without access to ground truth. Crucially, these statistics serve as a difficulty gate for adaptive compute allocation, dynamically routing problems across scaling regimes. First, Intrinsic Selection (iS) ranks candidates post-hoc, matching consensus-based algorithms across three domains and improving engineering design selection by 20% over pass@1 baselines. Second, Intrinsic Particle Filtering (iPF) generalizes this to step-level resampling, guiding generation toward high-confidence reasoning trajectories to improve pass@1 by 6.1 points on average on hard math problems. Finally, Particle Distillation (dPF) injects privileged guidance via early logit blending and KL-guided resampling, steering generation past systematic reasoning errors to satisfy expert rubrics, yielding up to 26.5% gains on complex clinical responses. Our pipeline applies seamlessly across broad-purpose, domain-specialized, and multimodal architectures, successfully extending ITS to open-ended domains without requiring trained reward models or exact ground-truth verification.

Giorgio Giannone, Mustafa Eyceoz, Shabana Baig +5

AI Innovation, Red Hat · DeCoDE Lab, MIT · Core AI, IBM

Particle Filtering Inference-Time Scaling Inference-Time Compute Allocation Inference-Time Guidance

Jun 4, 2026cs.CL

When to Think Deeply: Inhibitory Deliberation for LLM Reasoning

Reasoning Large Language Models can improve problem-solving performance through deliberative inference, but invoking slow reasoning for every input is computationally expensive and often unnecessary. We propose IDPR, a framework for response-conditioned inhibitory deliberation. IDPR first generates a concise intuitive answer and then uses an inhibition controller to decide whether that specific response should be released or suppressed in favor of slow reasoning. Unlike input-only routers, the inhibition controller conditions on the fast answer and fast-side evidence, including confidence, logit margin, parseability, and generation cost. We train the controller from paired fast-slow outcomes and select the inhibition threshold on a held-out validation set under an accuracy-first slow-call budget. On a held-out 5,000-example mathematical reasoning test set, IDPR invokes slow reasoning on only 8.20% of examples and improves accuracy from 47.90% to 48.92%. Under the same slow-call budget, random routing decreases accuracy to 46.76%, while the strongest confidence-based baseline reaches 48.22%. IDPR also achieves the highest corrective precision, showing that response-conditioned inhibition better identifies fast answers that benefit from slow reasoning.

Zhixuan He, Yue Feng

University of Birmingham, United Kingdom

Efficient Language Model Reasoning Inference-Time Compute Allocation LLM Routing Efficient Language Model Inference LLM Mathematical Reasoning

Jun 3, 2026cs.AI

Not All Errors Are Equal: Consequence-Aware Reasoning Compute Allocation

Test-time compute has emerged as an effective paradigm for improving large language model capability at inference time. Existing allocation strategies primarily prioritize tasks according to difficulty, uncertainty, or expected performance gain, implicitly treating prediction errors as equally costly. This assumption is often misaligned with real deployment, where failures can differ substantially in their downstream tasks. To address this limitation, this paper introduces consequence-aware test-time compute allocation by formulating a cost-weighted scheduling problem where the priority of a task is its failure consequence with the marginal gain of additional compute. In practice, however, marginal gain is difficult to predict before execution, so we propose a deployable scheduler that uses consequence as the routing signal. The scheduler predicts task consequence from pre-solution inputs and allocates the available premium compute to the corresponding top-ranked tasks. Experiments on the SWE bench Lite show that consequence provides information beyond task difficulty and can be predicted before solving. Under a fixed compute budget, consequence-aware routing achieves the best high-consequence task success, while overall accuracy remains competitive. A controlled within-model experiment further confirms the same advantage when only inference attempts are reallocated.

Liang He, Jingbo Wen, Haoyu Wang +4

Shanghai Institute of Optics and Fine Mechanics · The University of Sydney

Cost-Aware Inference Inference-Time Compute Allocation LLM Reasoning

Jun 2, 2026cs.AI

The Shadow Price of Reasoning: Economic Perspective on Optimal Budget Allocation for LLMs

Inference-time scaling has emerged as a critical avenue for enhancing Large Language Models' performance, yet real-world deployment is constrained by strict computational budgets. In this work, we formulate inference budget allocation as a global constrained optimization problem governed by economic principles. By modeling per-query reasoning utility with a shifted-surge function, we derive an optimal allocation policy based on a global shadow price that equilibrates marginal utility under resource scarcity. Based on this theory, we propose Constrained Latent-utility Equilibrium Allocation for Reasoning (CLEAR). It performs rational abandonment and reallocates resources from insolvent queries to solvable queries near their emergence thresholds. Extensive experiments on several reasoning tasks with different traffic streams demonstrate that CLEAR significantly improves the Pareto frontier of total token cost versus mean accuracy. In resource-scarce regimes, CLEAR achieves up to a 3x improvement in global accuracy compared to uniform allocation.

Xu Wan, Speed Zhu, Jianwei Cai +4

1Zhejiang University · 2Tecent HY Team · 3Peking University

Cost-Aware Inference Inference-Time Scaling Constrained Optimization Inference-Time Compute Allocation