Test Time

We study how model post-training and test-time inference design affect natural-language proof generation for hard olympiad mathematics. Starting from Nemotron 3 Ultra, we train two specialist checkpoints using supervised fine-tuning and reinforcement learning, and evaluate checkpoint choice, verification, and refinement. Based on these findings, we present an open-model test-time-compute pipeline. The system operates entirely in natural language, with no formal prover, external tools, or internet access. Three Nemotron 3 Ultra checkpoints - the general-availability model and two post-trained specialists - power an iterative search that generates, verifies, and refines candidate proofs; a separate high-compute stage then selects each final submission. The system scored 30 out of 42 points at IMO 2026, reaching the gold-medal threshold. We release the two post-trained checkpoints as well as the training data, the training and inference code, the submitted solutions, and Nemotron-IMO-Bench, a new benchmark of 200 novel olympiad-level problems.

Large language models (LLMs) are increasingly used for consequential decisions, making their explanations an important tool for auditing model behavior. Unfortunately, these explanations can be unfaithful, failing to reflect the actual reasoning underlying the model's decisions. We consider a setting in which an LLM provides both an answer and an explanation in response to a question. We identify two distinct dimensions of unfaithful explanations: incompleteness, meaning that the explanation omits factors that influence the answer, and unsoundness, meaning that the explanation cites factors that did not influence the model's answer. Existing approaches to improving LLM faithfulness include training-time methods, which require access to model weights and extensive computational resources, and test-time methods that largely focus on addressing unsoundness. We introduce a test-time approach that directly targets incompleteness. We remove from the input the concepts not credited in the model's explanation and re-query the model on the reduced input. This eliminates unmentioned influences while preserving the influence of mentioned concepts. Across two datasets, multiple model families, and two independent faithfulness metrics, our approach improves explanation faithfulness compared to both standard prompting and prompting to encourage faithfulness. Our method is model-agnostic and can be applied at inference time without modifying model parameters, providing a flexible mechanism for reducing hidden influences and improving the reliability and safety of LLM-assisted decision making.

We formulate indirect prompt injection as a test-time search over a task-dependent attack surface induced by the environment, user task, and injection task. To operationalize this formulation, we introduce an agentic attacker with a dedicated search harness that performs environment reconnaissance, structured reasoning over attack strategies, and adaptive evaluation using victim-agent feedback. Across heterogeneous tasks, we find that increasing attacker test-time compute improves vulnerability discovery and exploitation, while ablations show that explicit strategy management is important for avoiding redundant search and sustaining gains at larger budgets. These results suggest that agentic security evaluations should characterize both the attacker's search procedure and compute budget, rather than treating attack success as a budget-independent property of the victim. More broadly, our findings identify the attacker's adaptive search over the system attack surfaces as an important and underexplored security risk for tool-using agents.

Large Language Models (LLMs) often help users generate artifacts through iterative cycles of generation and revision in conversation. A challenge here is that, when users specify only a local change during revision, LLMs must instead identify the relevant dependencies and propagate the revision to all affected parts of the artifact. This paper studies this ability of LLMs on conversationally generated artifacts, where the artifact context and its dependencies may be embedded in the conversation history. Toward practical use, we also explore cost-effective test-time compute for this new setting. Specifically, we introduce a new benchmark for this setting, and evaluate nine revision methods, including sequential reflection and parallel sampling variants, using gpt-oss-20b/120b, gpt-5.4-mini, and qwen3.5-9b/27b/122b on the benchmark. The results show that baselines achieve accuracies of 68.3--93%, and the most cost-effective method is selecting from three parallel samples using either LLM-based or medoid selection, which improves accuracy by 2.2--9.7%. Our code and dataset are available at https://github.com/ntt-dkiku/llm-revision-propagation.

Large language models can often generate plausible mathematical reasoning traces, but reliably identifying the correct solution among multiple candidates remains a key challenge. Existing test-time reasoning pipelines typically rely on text-based verifiers that re-read each generated solution, making verification an expensive component of inference. Prior work has shown, however, that LLMs often encode correctness-related signals in their internal representations, including awareness of when their own answers are likely to be wrong. Building on this observation, we introduce HSRM, a lightweight hidden-state reward model that verifies candidate solutions by directly reading the generator's internal representations rather than re-processing its text. HSRM extracts hidden states from a frozen generator at reasoning-step boundaries and uses a small Transformer encoder to rank candidates. It is trained from self-generated trajectories with outcome labels, requiring neither human-written process supervision nor a large pretrained verifier. Across four mathematical reasoning benchmarks, HSRM matches or outperforms a 55M-parameter text-only energy verifier in 15 of 16 generator--dataset settings while using only about 2M parameters, providing an efficient alternative to text-only verification by reusing representations already computed during generation.

We propose claim-level falsification as a principle for test-time scaling and instantiate it through Claim-Level Reliability Assessment (CLR), a training-free framework that reallocates test-time compute from additional solution sampling to targeted verification. Since whole-trace evaluation often obscures decisive errors due to signal dilution from routine tokens, CLR condenses each reasoning trace into a compact set of decision-critical claims, thereby isolating its logical anchors. Furthermore, recognizing the inherent difficulty of generating entirely correct solutions under fixed model capabilities, CLR shifts the focus to semantic falsification. This approach exploits a fundamental asymmetry between solution construction and claim refutation. Constructing a valid solution requires a flawless reasoning path, whereas refuting an incorrect claim requires identifying only a single decisive flaw. This targeted search for negative evidence systematically compresses the survival space of high-confidence incorrect traces, effectively suppressing erroneous consensus via nonlinear reliability scoring. Across four LLMs and four reasoning benchmarks under matched budgets, CLR generally improves upon pass@1 and self-consistency. On GPT-OSS-20B/CMIMC25, for instance, CLR exceeds pass@1 by 27.15 percentage-points and raises self-consistency accuracy from 77.50% to 82.19% with 37.0% fewer tokens.

Test-time scaling often uses an external verifier, such as compilers and test cases in coding or trained value functions in robotics applications, to obtain high-quality rollouts. Verifier-free test-time scaling (or VF-TTS) is gaining extensive attention as a mechanism to enhance Large Language Model (LLM) reasoning, primarily because we do not have access to such high-quality verifiers in many real-world applications. Among existing VF-TTS methods, confidence-based VF-TTS methods, which compute and rank rollouts solely by confidence, are particularly promising. Such methods introduce near-zero overhead for sample evaluation and require minimal access to internal model states, making the methods highly flexible across models and tasks. In this paper, we demonstrate a critical limitation of existing confidence-based VF-TTS methods by showing that such methods catastrophically break down on complex tasks. We observe a very interesting phenomenon: uniformly high confidence frequently indicates a failure to explore, favoring confidently wrong answers. To address this, our core insight is that robust cognitive search requires a specific confidence trajectory pattern: such methods perform exploratory branching at the beginning, as manifested by low initial confidence, and converge to a high final confidence solution. To implement this insight, we introduce consilience, a novel selection framework that explicitly evaluates the temporal asymmetry of confidence in reasoning. We operationalize this via a combinatorial metric that actively penalizes high initial confidence while strictly demanding final certainty. Extensive experiments covering both graduate-level mathematics problems and free-form code generation demonstrate that consilience effectively outperforms existing baselines, validating our novel perspective on completion confidence.

We independently reproduce two recent methods for making large language model (LLM) reasoning more reliable, and stress-test them across domains and models (RPC across four new task domains with Qwen3-8B, LCF across four 7-8B models). The first, RPC, aggregates token probabilities and self-consistency at inference; the second, LCF, trains projectors that split hidden states into "content" and "logic" and edits the logic part toward a valid region. Validating such reliability claims matters because the original evaluations are run by each method's own authors and were never independently reproduced or stress-tested across models and domains, and LCF shipped no public code. We re-run RPC's published-path aggregation and re-implement LCF's projector, contrastive, and intervention pipeline, then extend both to text-to-SQL, legal extraction, fallacy identification, and precedent grading, and probe LCF's representation directly. RPC reproduces the original grid exactly on the authors' released reasoning paths; on four new domains its edge over self-consistency is never significant (ties or small mixed differences, paired p >= 0.28), and on BIRD, the one domain where we vary the budget, the edge grows with K as predicted but its largest gap (+2.5 accuracy at K=32, p=0.16) reverses to -0.25 when we enlarge the sample to n=200. LCF's logic-validity direction is real but weak (0.82 separability at the single best sub-layer versus 0.95 for a semantic-attribute control); its one positive effect (Qwen3 $Δ$Prob) is not significant (p=0.56), while it significantly reduces $Δ$Prob on two of the other three models.

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.

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.

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.

Test-time scaling improves LLM reasoning by using additional inference compute, but wider sampling alone can suffer from diminishing returns: new rollouts often repeat existing answer patterns instead of adding useful reasoning diversity. Verifier-based selection offers an alternative, but its performance depends on the calibration of an external reward model. We propose a verifier-free breadth--depth refinement framework that uses test-time compute to both explore and improve candidate solutions. The method samples multiple independent reasoning rollouts, refines each rollout through iterative self-critique and self-correction, and aggregates the refined answers by majority voting. Breadth preserves diverse initial attempts, while depth repairs local reasoning errors before aggregation. Across AIME24, AIME25, AMC, OlympiadBench, and MATH500, our method consistently improves over greedy decoding, majority voting, verifier-based best-of-$N$, beam search, and lookahead decoding across multiple open-weight models. For instance, with Qwen2.5-1.5B, accuracy increases from the strongest verifier-based baseline to $58.0\%$ on MATH500, and from $25.0\%$ to $32.5\%$ on AMC. These results show that test-time compute can be more effective when used to refine sampled trajectories rather than only to sample more candidates or rely on verifier-guided selection.

As chart images, tabular data, and visualization code play increasingly important roles across diverse domains, cross-representation understanding across these modalities poses fundamental challenges for AI systems: the relationships across representations are inherently \textit{one-to-many}, supervision is ambiguous and costly, and model optimization lacks a principled signal that is both direction-adaptive and representation-generalizable beyond task-specific objectives. We introduce CoCoEvolve to improve consistency across chart, table, and code representations. Instead of treating cross-representation mapping as a one-to-many problem, we define explicit one-to-one correspondences and optimize models using agreement between representations, without additional annotations. During training, CoCoEvolve@Train performs co-evolution across the chart-table-code cycle, while CoCoEvolve@Test applies the same consistency objective at inference time for test-time co-optimization. We also present CoCoEvolve@Eval, an evaluation suite covering all six cross-representation tasks. Across four benchmarks, CoCoEvolve improves performance in both training-time and test-time settings. Our project page: https://xhguo7.github.io/CoCoEvolve/.

Test-time scaling improves LLM reasoning by generating and aggregating multiple candidate answers, yet many pipelines use fixed per-query budgets that spend the same compute on easy and difficult prompts. These fixed budgets are also difficult to inspect because they do not explain why a given prompt receives a particular number of samples. We propose adaptive} test-time scaling with a lightweight fuzzy controller that maps interpretable signals, including estimated prompt complexity and model confidence, to a per-query sampling budget. The controller assigns fewer samples to easier or more confident prompts and more samples to harder or less certain prompts, making inference-time compute inspectable rather than fixed or opaque. We evaluate under a fair-alignment protocol with matched decoding settings and controlled answer selection, and compare against best-of-$N$, compute-aware scaling, and self-certainty-based baselines on question-answering and mathematical reasoning tasks. Across models and datasets, adaptive fuzzy control improves over several standard baselines and remains close to a selector-matched full-budget control while reducing the average number of samples. These findings suggest that interpretable adaptive sampling is a practical direction for more efficient test-time reasoning in large language models.

Post-training reinforcement learning (RL) algorithms are commonly used to align large vision-language models (LVLMs) with human intent and the requirements of visual reasoning tasks. However, existing RL-based alignment methods are often resource-intensive and encounter mismatches between training objectives and inference-time distributions. To bridge this gap, we propose a novel test-time alignment approach that leverages trajectory-guided structured sampling for dynamic inference-time refinement, achieving better alignment with visual grounding and ensuring logical consistency. Our approach begins with curating a reasoning memory bank via a trajectory learning algorithm, which decomposes complex question solving into ordered sequences of predefined reasoning patterns. It subsequently accomplishes inference-time alignment by first collecting trajectories from reasoning memory bank to establish a global structural reasoning prior, and then using an iterative Markov Chain Monte Carlo (MCMC) algorithm for localized multi-objective refinement of the reasoning trace. Experiments across multiple multimodal reasoning datasets demonstrate that our approach significantly improves accuracy without incurring prohibitive inference overhead. These results establish trajectory-guided test-time sampling as a scalable and effective alternative to traditional post-training alignment, particularly for complex visual reasoning tasks.

Test-time scaling lifts large language model reasoning by sampling many candidate solutions and selecting among them, yet the same recipe transfers poorly to vision-language models (VLMs): recent work shows that simple majority voting beats selection methods built on the model's own self-verification, apparently because at the selection layer an image-grounded answer and a confident guess from the language prior look the same. A natural fix is to make the selection signal one that cannot be computed without the image. We study Perturbation Grounded Selection (Pgs), a label-free, training-free rule that scores each candidate by whether the model re-derives it under label-preserving perturbations of the input (cropping, background masking, mild photometric or geometric jitter); Pgs recovers majority voting when the perturbation set is empty. The decisive question is not whether Pgs beats chain-of-thought only majority voting, but whether the perturbation term adds anything once decoding format and budget are controlled. We therefore introduce a format-matched control (MatchedCtrl): the same short, no-CoT draws spent on the original image. Across TextVQA, MATH-Vision, MMMU, and ViLP, with a Qwen headline (three-seed means) and LLaVA-OneVision coverage in matched-budget selector tables, Pgs appears to beat plain majority voting by up to +31.8 points on TextVQA (Qwen), but MatchedCtrl tracks or exceeds Pgs within noise on every benchmark, including the vision-required ViLP; no Qwen category shows a significant gain over this control. The stability gap is real and image-dependent (up to +0.48), yet does not predict per-instance wins. The result is negative and diagnostic: perturbation consistency is at best a partial diagnostic of visual dependence and, on its own, not a usable selection signal once format is controlled; gains reported against CoT-only majority voting overstate such methods.

In animation production, paint-bucket colourisation for hand-drawn animation is a labour-intensive procedure that assigns each enclosed region in line sketches a colour from reference design sheets. Recent automatic paint-bucket colourisation pipelines mirror this workflow via region correspondence, but correspondences can be brittle when regions are ambiguous fragments without proper context. In this paper, we propose Palette Context Assisted (PeCA), a new training-free, plug-and-play framework for animation video colourisation that aims to close this gap at test-time via reasoning over spatial and temporal contexts. Extensive experiments on existing benchmarks and a newly introduced long-video test case show consistent performance boosts.

Scaling test-time computation can improve language-model reasoning, but uniform budgets waste computation on easy inputs, while verifier-guided refinement relies on external feedback. We introduce Self-Verifying Refinement (SVR), an oracle-free multi-turn reinforcement learning framework that learns to use self-verification as a compute-control policy. At each turn, the model produces a solution together with a discrete correctness verdict and a confidence score; it retains the current answer only when the verdict is Correct and confidence exceeds a threshold, and otherwise continues refinement using its own self-verification. Ground-truth correctness is used only to construct training rewards and is never exposed to the policy through refinement prompts or required at inference. SVR is trained with GRPO on fixed-horizon trajectories using rewards that promote solution correctness, calibration-aware self-verification, and stop-ready correct states; adaptive stopping is activated only at inference. On seven mathematical reasoning benchmarks with Qwen3.5-2B, SVR achieves a macro-average accuracy of 0.563 with only 2.99 inference turns on average. In the evaluated complete-system comparison, it exceeds standard GRPO, strong multi-turn baselines, and a fixed-budget oracle-guided score-feedback reference while requiring substantially fewer turns than fixed ten-turn inference. These results demonstrate that learned self-verification can serve as an effective internal control signal for answer retention and adaptive test-time compute allocation.

Test-time search lets small video diffusion models rival larger ones, but costs 2-10x more. All candidates are fully denoised, although most are discarded. Training-free caching makes each rollout 2-3x faster at near-lossless quality. Composition is safe only if lossy caching preserves verifier rankings. We present the first study of whether caching corrupts candidate ranking in video test-time search. On Wan2.1-T2V-1.3B with an adaptive caching wrapper (~2x per-candidate speedup), ImageReward scores seed-matched cached and full rollouts. Median per-prompt Spearman rank correlation is 0.905, with 72% top-1 agreement on the VBench suite. VBench-2.0 replicates this result on a harder suite. Recomputing the cached winner at full compute retains 90-94% of the full-search gain. Errors cluster among near-tied candidates, making corruption self-limiting. This finding leads to CachedSearch. It explores every candidate with aggressive caching, then re-generates only the winner at full compute. At N=8, it captures 94.7% of best-of-N's gain at 63% of the cost. Capture rises with width. At matched budget, it searches twice as wide for 38% more gain. The result holds from 1.3B-14B across six models and four families: Wan, LTX, CogVideoX, and Hunyuan. Wan2.1-14B matches the 1.3B model's fidelity. Mid-trajectory pruning multiplies the exploration saving to 3.11x at 88.6% capture. Ports to other model families require recalibrating a single parameter, showing that fidelity tracks architecture rather than parameter count. CachedSearch is training-free, verifier-agnostic, and orthogonal to the search algorithm, making it a plug-in multiplier for test-time scaling.

When does a weight-tied looped transformer -- one block applied T times -- implement an actual algorithm? We answer with four findings from controlled populations on group word problems. (1) The budget law: free training installs a linear computation frontier, a mechanism that solves v positions per loop, whose speed is priced by the training contract: v ~ n_train/T_train (exponent 0.98 +/- 0.04, R^2=0.99), exactly unity under T=n training. SGD selects a frontier matching the minimum the contract demands; granting more test-time loops than ever trained rescues late positions at fixed input length, yielding a principled halting rule T* = ceil(n / v-hat). (2) Architecture prior, not expressivity, picks the algorithm: standard-depth transformers learn parallel scans on this family; weight tying flips the selection to the serial frontier, even when positional addressing for a log-depth scan is supplied. At matched depth and parameters, untied models extrapolate worst and fail to learn A5 at all. (3) The walls are not where circuit complexity says: NC1-completeness costs nothing (A5 generalizes fully), while group order does (S5's 120x120 operator deadlocks joint learning) -- and an operator-first curriculum dissolves the wall in every seed. (4) Mechanisms are portable, not mandatable: warm-starting across budget contracts transfers the algorithm in every seed, re-pricing its speed, while imposing seriality through the input schedule fails where free training succeeds. These results are invisible to standard instruments, which provably saturate at the fixed points trained loops converge to. We introduce a head instrument, the convergence-time scaling tau(n,i), validate it causally via damage cones whose slope reproduces v, and show in-distribution head measurements predict out-of-distribution fate where tail metrics do not. Results replicate on the public easy-to-hard benchmark.

Attention-based models often develop attention sinks, where a small number of tokens repeatedly attract attention and accumulate unusually large activations. In vision transformers, these outliers are closely related to registers, which have been diagnostically linked to global, low-frequency image structure. Existing work has largely studied registers through interpretability analyses and linear probes, leaving open whether they can be operationalized as plug-and-play signals for generation without retraining. We revisit this question in tokenized image generation. Using OpenCLIP and DINOv2 on ImageNet, we find that test-time register features exhibit stronger low-frequency concentration than both [CLS] readouts and patch-mean features, and show a consistent (albeit moderate) correlation with pixel-space DCT low-frequency energy. Motivated by these diagnostics, we introduce RegToken, a training-free procedure that converts register structure into a small set of global prior tokens by (i) NFN-based layer localization, (ii) TokenRank-guided subspace extraction, and (iii) a projection-and-conservation update on the register subspace. Inserted into a frozen compact 1D token generation pipeline, RegToken improves ImageNet generation and alignment metrics (e.g., FID-5k 20.5 to 20.1, SigLIP 3.6 to 3.9) without modifying pretrained weights, and accelerates test-time optimization (Steps@$τ$ 74 to 52). Overall, our results suggest that structures often viewed as attention artifacts can be repurposed as lightweight global priors for tokenized generation.

Understanding long videos with multimodal large language models (MLLMs) requires selecting a compact set of frames from thousands of candidates, yet identifying the right frames seemingly requires understanding the video first. We resolve this circular dependency with a simple observation: cross-modal attention at validation-selected extraction layers in MLLMs already provides query-relevant frame evidence without requiring autoregressive generation. We exploit this property to build DAFS (Dynamic Attention-based Budget-aware Frame Selection), a training-free frame selector. A lightweight MLLM selector, even with only 2B parameters, can extract frame-level evidence by converting selected-layer attention into relevance scores through query-conditioned aggregation. This enables cross-frame comparison without autoregressive decoding. To handle the selector's own context constraint, we formulate the joint allocation of candidate pool size and per-frame token budget as a discrete optimization problem solved by dynamic programming. Under a 32-frame budget, our selector improves over uniform sampling by up to 6.4 points on Video-MME and outperforms prior training-based selectors under matched frame budgets, while generalizing across selector and answerer backbones, and across tasks, without retraining.

Automating analog circuit topology design is essential to reduce the extensive manual effort required to meet increasingly diverse and customized application demands. Recent advances have applied sequence-to-sequence fine-tuning on pretrained language models to directly generate circuit topologies from user specifications in a single pass. However, these one-shot generation methods failed to generate complex circuits due to their exponentially growing search spaces and limited training datasets. In this paper, we present EXPLORE, a search-enhanced framework that integrates simulator-guided Monte Carlo Tree Search (MCTS) with transformer-based decoding to enable test-time scaling for analog topology generation. By leveraging language-model priors and bypassing high-confidence structural tokens, EXPLORE allocates expensive simulator budget primarily toward topology-altering decisions during search. On a 6-component benchmark at a tight tolerance of 0.01, EXPLORE raises the success rate from 12% for one-shot generation and 33% for a sampling-and-filter baseline to 65%, and lowers MSE by over 20% relative to sampling-and-filter under the same search budget. These results establish EXPLORE as the first framework to integrate structured test-time search with LM decoding for analog topology generation, and a practical step toward scaling LLM-driven design automation.

There are two standard ways to spend more compute at test time: let a model reason longer, or sample more attempts and keep one. Both share a hidden limit: they are internal. Every extra token comes from the same frozen weights and the same prompt, so neither can tell the model anything it does not already know. We study a third way, interaction: the model proposes an artifact, an external instrument observes how it actually behaves, and the model revises. Each cycle imports a real observation, so interaction breaks through the ceiling the other two hit. We argue that a single variable governs this third axis, grounding, and that it must hold on both sides of the loop. The feedback that drives revision must come from an instrument that actually observes the flaw, and so must the metric that scores the result. On hard coding tasks at a fixed token budget, reasoning-only and best-of-N sampling both plateau (the latter even when an oracle picks the best sample), while every interaction strategy keeps improving; our proposer-reviewer harness reaches a perfect 100% pass rate with no run-to-run variance, and the gain holds across three model families. On rendered visual artifacts, the usual judge (a vision-language model, or VLM, reading a screenshot) rates 14 of 15 visibly broken figures "perfect," because the screenshot hides the flaws before the judge can see them. A tool that measures the real layout instead shows the loop removing 40-74% of defects across four modalities; and that same VLM, used as the reviewer, makes slide layouts worse where the measuring tool repairs them. Interaction scaling is real and distinct from reasoning and sampling, but only visible when both the feedback and the metric are grounded.

Despite their central role in fault detection, test oracles remain challenging to construct effectively. Recent learning based methods address this challenge by automatically generating test assertions, yet even if syntactically correct, they are often ineffective in revealing bugs. Rather than generating assertions, this study explores a different approach by training a model to directly predict whether a given test prefix passes or fails. We present FOCAL, an emerging code LLM-based discriminative oracle predictor. It learns from labeled pairs of test prefixes and methods under test, employs losses that emphasize failing cases during training, and grounds its predictions in statement level behavioral evidence. Compared with the baseline method SEER, we substantially improve performance on failing cases for unseen projects and provide richer explanations. A preliminary evaluation on fault-detection benchmarks and automated test-generation artifacts shows that our approach is highly accurate within its training distribution and substantially improves failure detection on previously unseen projects where prior discriminative oracles collapse. Moreover, the highlighted statements are supported by behavioral explanation checks. These early results suggest that fail-aware discriminative oracle prediction can complement existing approaches such as fuzzing, search-based testing, and LLM-based test generation. These techniques produce test prefixes at scale but often lack fault oriented oracles. In future work, FOCAL could take generated test prefixes and attach fault-aware predicted oracles to them, turning high-volume input generation into executable tests that are more likely to expose semantic failures.

The behavior of an LLM agent is determined not only by the underlying model, but also by its harness: the executable program that constructs context, invokes tools, verifies intermediate results, and recovers from failures. Existing approaches optimize such harnesses before deployment, searching training or development data for a fixed agent workflow that is then frozen at test time. This limits adaptation when the test distribution, failure modes, or tool interactions differ from those seen during development. We ask whether the harness can instead be optimized during evaluation itself, using only the unlabeled execution traces the agent produces on the test inputs. We introduce Test-Time Harness Evolution (TTHE), which treats the executable harness as the state of test-time adaptation. During evaluation, TTHE maintains a population of candidate harnesses and refines them through an agentic proposer that reasons over their execution traces, without gold labels or task-specific supervision; a judge then commits an improved harness from execution-derived proxy signals, and the selected program persists to govern subsequent inputs. Crucially, TTHE does not update model weights, require gold labels, or train a separate adaptation model: solver, proposers, and judge are different roles and harnesses around the same frozen LLM, so all adaptation occurs through changes to the surrounding program. Across text-to-SQL, competitive programming, software engineering, data-science coding, and agentic tool-use tasks, TTHE improves fixed ReAct-style baseline harnesses, yielding persistent, inspectable improvements rather than a pre-searched workflow or per-query retries. These results recast test-time adaptation for LLM agents as evolution over executable control programs and identify execution-derived proxy reliability as a central challenge for robust unsupervised agent improvement.

Large Language Models achieve strong code generation for high resource languages like Python and Java but suffer sharp performance drops on Low-Resource Programming Languages~(LRPLs) such as Julia. Improving Small Language Models~(SLMs) for these languages faces a trilemma: Supervised Fine-Tuning~(SFT) is bottlenecked by data scarcity, inference-time scaling is too expensive for deployment, and Reinforcement Learning from scratch yields near zero advantages. We propose a three-phase pipeline that resolves this trilemma by decoupling syntax acquisition from algorithmic reasoning. First, we \emph{left-shift} inference-time compute to an offline data synthesis engine that uses iterative compiler and test feedback to generate verified training examples. Second, we fine-tune an SLM on this synthetic, verified data to embed strong syntactic priors. Third, we apply Reinforcement Learning with Verifiable Reward~(RLVR) grounded by language-agnostic Input/Output tests, where the SFT prior constrains exploration away from syntax errors. Applied to Qwen3-8B, our pipeline improves pass@1 by up to +7.6 points on MultiPL-E and +14.2 points on the Agnostics LiveCodeBench for Julia compared to SOTA results. Furthermore, the pipeline only used $\frac{1}{3}$ data and $\frac{1}{6}$ cost over the previous state-of-the-art. We further demonstrate that the pipeline generalizes to Ballerina achieving 49.7% MultiPL-E Pass@1, a language with near-zero pretraining representation. Ablations confirm that both the SFT phase and execution-grounded rewards are necessary for stable training.

Steering robot foundation models (RFMs) toward new task variants or user-preferred behaviors remains challenging, often requiring additional robot demonstrations, task-specific fine-tuning, or long-context conditioning. We present WAM-TTT, a test-time training framework for steering world action models from raw human videos. Rather than treating human videos as trajectories to imitate, WAM-TTT absorbs them into a lightweight adaptive memory inside a frozen WAM through self-supervised video prediction. To make this memory useful for control, we introduce a meta-training stage that aligns human demonstrations with robot behaviors using paired human-robot data and a key--value memory reconstruction objective. At test time, only unlabeled human videos are required to adapt the memory, while the pretrained WAM remains frozen. This enables efficient and reusable steering without robot actions, human-side annotations, or task-specific fine-tuning, while preserving the generalization ability of the foundation model. Extensive experiments show that WAM-TTT consistently outperforms in-context human-video conditioning baselines across diverse manipulation tasks and generalization settings.

Recent progress on ARC-AGI-1 from disclosed architectures has come broadly from two regimes: heavy test-time compute over frontier models (evolutionary search, exhaustive sampling, extended chain-of-thought), or benchmark-specific training in which small models are fine-tuned on ARC data, often with task-specialized architectures. We study a third regime: an open-weight model in non-thinking mode (DeepSeek V3.2) under a strict budget, with no ARC-specific fine-tuning. We study what is recoverable through architecture alone, building agentic harnesses that decompose pattern-discovery and program-synthesis stages explicitly. First, we introduce an Explorer-Definer Pipeline that separates pattern discovery from executable transformation synthesis, implemented as a two-stage agent pipeline. Next, we present the Reflective Orchestrator, which augments the pipeline with autonomous exploration of new transformations when previous hypotheses fail on training pairs. On the ARC-AGI-1 public 400-task evaluation set, the pipeline reaches 57.50% pass@2 at $0.25 per task, and the orchestrator reaches 67.25% pass@2 at $0.62 per task. Together these architectures lift a 15.50% one-shot baseline by ~52 points without benchmark-specific training or heavy test-time compute. Furthermore, the orchestrator-driven lift tests a falsifiable diagnostic the pipeline produces; unbiased pass@k analysis suggests the pipeline is generation-bound, not selection-bound (selection via training-pair accuracy captures ~95% of the candidate ceiling) and predicts that significant improvement requires broader generation, not better ranking. The orchestrator implements this prediction via adaptive re-exploration and confirms it (unbiased pass@1 lift +9.81 pp, matching selection-mediated pass@2 lift). An additional pipeline ablation identifies its think tool as a significant component, with removal reducing pass@2 by 5.75 pp.

In the automotive domain, in-car scene understanding (ISU) enables the detection of safety-critical events, such as driver distraction, and supports drivers or passengers by analyzing the in-car scene and adapting the environment (e.g., ambient lighting). The industry is increasingly exploring vision-language models (VLMs) to interpret camera-recorded in-car scenes and extract information for downstream reasoning tasks. However, VLMs may generate incomplete, erroneous, or misleading scene descriptions, highlighting the need for systematic testing. Collecting real in-vehicle data is costly, difficult to scale, and often infeasible, particularly in early design stages. In this paper, we present ISU-Test, an automated testing approach that combines rendering-based scene generation with search-based testing to evaluate ISU systems. By framing testing as an optimization problem and systematically modifying scene parameters, our method generates diverse in-car scenarios and explores a wide range of configurations. We evaluate ISU-Test on both an industrial prototype and open-source VLMs across two case studies: question answering and captioning, comparing against randomized scenario generation. Results show that ISU-Test significantly outperforms the baseline, achieving up to 10 times higher failure rates and up to 3.6 times higher failure coverage.

Memory-based discrete vision-language navigation (VLN) agents must act under partial observability, yet even strong frozen backbones remain vulnerable at test time. Two common failure modes are stale historical evidence at memory readout and inefficient local backtracking during action selection. We present DART-VLN, a training-free test-time control framework for discrete VLN. DART-VLN combines Test-Time Memory Decay, a read-side memory reweighting rule that suppresses stale and redundant evidence without rewriting stored content, with Anti-Loop Regularization, a lightweight next-hop penalty that discourages immediate reversals during action selection. The framework introduces no new learnable parameters and leaves the learned backbone unchanged. Experiments on R2R and REVERIE show a consistent pattern: decay-only provides stable read-side gains, while decay+anti-loop achieves the best overall quality-efficiency trade-off, yielding shorter trajectories, lower runtime, and improved navigation performance in key settings. Behavioral analysis further confirms that anti-loop regularization reduces local backtracking and improves path efficiency under frozen backbones. Overall, the results show that modest test-time control can make memory-based discrete VLN more reliable and efficient without retraining.

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%.

Improving the reliability of large language models (LLMs) at inference time is a central challenge in structured reasoning tasks such as Text-to-SQL. Common test-time inference strategies, including Best-of-N sampling and Majority Voting, rely on heuristic signals such as execution success or output frequency, which provide limited semantic discrimination across candidate outputs. In this work, we study Outcome Reward Models (ORMs) as learned semantic scoring functions for test-time verification in Text-to-SQL. While ORMs have been previously explored for test-time scaling and alignment, their application to structured query generation remains underexplored. We introduce GradeSQL, a scalable framework for training task-specific ORMs via automated candidate generation and execution-based labeling, enabling verifier training without manual annotation. We integrate ORMs into a verification-driven Best-of-N pipeline and evaluate our approach on the BIRD and Spider benchmarks across multiple open-source LLM families. ORM-based selection consistently outperforms execution-based Best-of-N and Majority Voting, with gains of up to +4.33% on BIRD and +2.10% on Spider. We further show that ORMs scale effectively with larger candidate sets and yield stronger improvements on complex queries. Overall, our results demonstrate that ORM-based verification provides a simple, effective, and scalable alternative to heuristic test-time selection strategies for Text-to-SQL. Code datasets and models are publicly available.

Time-series deployments often need delayed feasible decisions, not only accurate forecasts. UC-Search is a trace-only retained-search layer for delayed constrained control: a frozen backbone emits forecasts or action scores, a hard-feasibility automaton rolls paths forward, and bounded search returns the first action of a feasible trajectory. The main claim is conditional: retained lookahead can improve delayed constrained decisions only when delayed feasible-set coupling, retained-prefix premises, and fail-closed release certificates hold. The promoted public endpoint is Phase128 certified M4 expanded40: validation selects Certificate-Constrained Retained Pareto Beam with $λ=0.25$, the held-out test has certificate/risk-active rates $1.0000/0.9642$, and the weakest family remains above the unchanged $0.95$ gate at $0.9516$ on M4Weekly. The author-defined public $9$-family suite remains an uncertified stress-test boundary. The paper reports a trace-only mechanism, one certified public endpoint, failed-route certificates, and deployment boundaries rather than a universal risk-control theorem.

Text-conditioned motion generation is a promising interface for programming humanoid robots, yet current generators are often trained on human motion datasets retargeted to robot morphologies. Although such data provides rich semantic and kinematic priors, it fails to capture the nuances of whole-body tracking controllers, including balance, contact dynamics, actuation limits, and controller-specific failure modes. As a result, generated motions can be semantically plausible but difficult or impossible for the robot to execute. We introduce TEXEDO, a test-time scaling framework for humanoid motion generation that improves motion quality without requiring a stronger underlying generator. Given a text prompt, TEXEDO samples multiple candidate motions from a pretrained text-conditioned generator and selects the best motion that is both executable and task-aligned. The reward model combines a dynamic feasibility verifier, distilled from whole-body tracking rollouts to predict physical executability, with a semantic alignment verifier that measures text-motion alignment in a learned co-embedding space. Our pipeline treats dynamic feasibility as a hard constraint and semantic alignment as the selection objective within the feasible set. Through large-scale simulation studies and real-world deployment on a Unitree G1 humanoid robot, we show that TEXEDO consistently improves both tracking fidelity and text alignment. These results demonstrate that grounded verification is an effective path toward deployable language-guided humanoid motion generation. Project website: https://jianuocao.github.io/TEXEDO/

As multimodal agents move from interface understanding to real software control, successful trajectory discovery in live desktop environments becomes a key challenge. GUI tasks require long-horizon sequences of precise mouse and keyboard actions, while feedback is sparse, delayed, and costly to obtain through VM rollouts. We propose Environment-Native Verified Search (ENVS), a training-time search-and-filter pipeline that uses the environment to construct verified supervision before policy optimization: it branches over behaviorally distinct GUI actions in live OSWorld VMs, verifies successful leaves, and trains from globally balanced step-level supervision. To evaluate robustness under realistic desktop interruptions, we also introduce OSWorld-Noisy, a dynamic benchmark for recoverable desktop interruptions that preserves the original tasks while testing whether agents can refocus, dismiss, wait, or recover under live perturbations. On the 300-task OSWorld pool, ENVS reaches 30.3 pass@8 on original evaluations and 29.0 on OSWorld-Noisy, outperforming matched ARPO-style online RL while reducing compute from 184-192 to 138-153 GPU-hours; even with only 30% of its search data, ENVS reaches 27.0 pass@8, exceeding ARPO from the base model. Training from noisy environments also better preserves visual-reasoning abilities on auxiliary benchmarks, including OSWorld-G Refusal (16.7 vs. 1.9) and BLINK Functional Correspondence (26.2 vs. 23.1).

Pre-trained foundation models have demonstrated remarkable success in many domains, enabling a unified backbone to generalize across diverse downstream tasks. However, extending this paradigm to graph learning remains challenging due to the intrinsic mismatch between graph data and fixed architectural designs. In this work, we show that this limitation can be overcome via recurrent graph models. To achieve this, we conduct a systematic theoretical analysis, rigorously deriving step dependence as a necessary and sufficient condition for an adaptively convergent recurrent process. Building on this foundation, we propose AdaR, an Adaptive Recurrent graph model, empowering flexible test-time computing on various downstream tasks without changing model parameters. To enable adaptive inference, AdaR explicitly encodes normalized step information and representation-target relations into the recurrent updates. To ensure convergence of the recurrent process, AdaR employs gradient-based supervision signals that guide representation updates throughout the recurrence. Empirical results demonstrate that AdaR consistently outperforms strong baselines in both inductive and transductive settings.

This paper introduces the CERTH-ITI solution for the MediaEval NewsImages 2026 challenge, which focuses on generating images related to news headlines. Inspired by the Actor-Critic paradigm in reinforcement learning, we present a test-time, model-agnostic Actor-Critic Image Generation approach (ACIG). ACIG generates prompts for image creation, produces the images, evaluates the generated results, and if needed refines the image generation prompts accordingly in a feedback loop. ACIG achieved the best results in the NewsImages 2026 challenge, according to the challenge's leaderboard.

Test-time scaling via sequential revision has emerged as a powerful paradigm for enhancing Large Language Model (LLM) reasoning. However, standard post-training methods primarily optimize single-shot objectives, creating a fundamental misalignment with multi-step inference dynamics. While recent work treats this as multi-turn reinforcement learning (RL), conventional approaches optimize over the multi-step trajectories directly, failing to further exploit the high-quality mistakes in intermediate steps that model can learn from correcting them. We propose a two-stage iterative framework that alternates between online data/prompt augmentation and policy optimization. By converting the intermediate steps (``near-miss'' answers) in the successful recovery trajectories into decoupled revision and verification prompts, our approach concentrates training on both effective answer transformation and error identification. This approach enables efficient off-policy data generation and reduces the computational overhead of long-horizon sampling compared to standard multi-turn RL. On LiveCodeBench, using publicly available test cases as feedback, we observe gains of +6.5 points over the RL baseline and +4.0 points over standard multi-turn training. Beyond coding, our approach matches the previously reported SOTA result on circle packing while using the smallest base model (4B) and far fewer rollouts than the much larger evolutionary search systems. Math results under ground-truth verification further confirm improved correction ability. It also generalizes to out-of-distribution constraint-satisfaction puzzles such as n_queens and mini_sudoku, where correctness is defined entirely by problem constraints. Code is available at https://github.com/yxliu02/REVES.git.

Quantum chemistry simulations underpin modern materials discovery, yet their impact is limited by steep computational cost and dependence on fixed approximation schemes. Foundation models, such as machine-learned interatomic potentials, have accelerated parts of this workflow, but their reliance on large-scale pretraining restricts adaptability at the frontier of chemical space, where methodological innovation and sparse data are the norm. Agentic AI systems can automate existing simulation pipelines, yet they remain constrained by the predefined tools and algorithms they orchestrate. In response, we introduce LADeQ, an LLM-guided workflow that discovers, implements, and benchmarks candidate approximation algorithms at test-time within existing quantum chemistry codes. Rather than selecting from a predefined repertoire, LADeQ constructs candidate approximation schemes on demand, drawing on techniques from disciplines such as spatial statistics, circuit simulation, and kernel methods that have had little prior presence in electronic-structure theory. Because it builds on an out-of-the-box language model, LADeQ requires no task-specific pretraining or curated data, and the resulting implementations are transparent and inspectable, with explicitly traceable approximation errors that enable principled control of accuracy--efficiency trade-offs. We show that LADeQ accelerates coupled cluster singles and doubles (CCSD) and configuration interaction singles and doubles (CISD) calculations while keeping correlation-energy errors within user-specified tolerances, demonstrating autonomous, objective-driven discovery of approximation algorithms inside existing electronic-structure solvers.

Extending a vision-language-action (VLA) policy to a new task typically requires task-specific teleoperated demonstrations and per-task fine-tuning, making adaptation costly in both data collection and compute. In this paper, we show that this target-side per-task adaptation cost can be replaced by retrieval. Our retrieval-augmented policy is trained once on paired demonstrations from the target embodiment (query) and a cheaper embodiment (pool, e.g., human-hand video), then frozen. New tasks are added at deployment by appending pool-side demonstrations to a retrieval pool. The frozen policy conditions on retrieved trajectories at every control step, so new tasks are absorbed by indexing data rather than updating parameters. Fine-tuning is needed only to take on a new, unseen embodiment, not for each new task. We show that retrieval improves policies beyond a specific backbone, including standard VLA policies, but its effect is especially pronounced in Cosmos Policy, a video-generation-based world-action model (WAM). In this setting, retrieval supplies coarse task progression, while the WAM's future-image objective provides an additional visual consistency signal that strengthens the retrieval-conditioned actions. On PushT, we study how retrieval provides a reusable high-level motion prior for cross-embodiment generalization to unseen goal angles, while on RoboTwin 2.0 our method outperforms cross-embodiment baselines on unseen tasks, and we additionally demonstrate the method on a real robot.

Collaborative generation, which coordinates multiple diffusion trajectories to extend the capabilities of pretrained priors, has emerged as a powerful paradigm for extending the applicability of diffusion models. Among existing approaches, diffusion synchronization provides a scenario-agnostic solution by introducing general guidance mechanisms. However, current synchronization approaches rely heavily on heuristics and still require task-specific tailoring, which limits their generalizability and performance. In this work, we mathematically derive a synchronization framework based on optimal control, providing a principled explanation of diffusion synchronization. During sampling, we optimize control variables to guide multiple trajectories toward coherent solutions while remaining close to the underlying diffusion prior. Our method operates entirely at test-time without additional training, thereby enabling broad applicability across diverse generation scenarios when combined with strong pretrained priors. We demonstrate consistent improvements over baselines on three representative collaborative generation tasks, covering a wide range of modalities and applications. Beyond performance gains, our work establishes a novel foundation for collaborative generation, opening a principled path toward extending pretrained generative models to new collaborative generation settings.

While real-world applications of reinforcement learning (RL) are becoming increasingly popular, the security of RL systems deserve more attention and exploration. In particular, recent work has revealed that RL agents are vulnerable to backdoor attacks, where a victim agent behaves normally under standard conditions but executes malicious actions when a specific trigger is activated. Existing backdoor defenses for RL either require access to the agent's internal parameters, operate only at the model or trajectory level, or are limited to specific attack types. To ensure the security of RL agents, we propose \texttt{PolicyGuard}, a \textit{test-time step-level} backdoor defense which leverages Gaussian Process (GP) posterior variance and adapts pseudo trajectories to enable uncertainty computation for individual time step. Besides, we also provide theoretical foundations to explain the efficacy of GP posterior variance. Extensive experiments across seven RL games demonstrate that PolicyGuard achieves state-of-the-art detection performance in most cases, with average AUROC of 0.856 for perturbation-based attacks and 0.859 for adversary-agent attacks.

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/.

Test-time compute (TTC) scaling has emerged as a powerful paradigm for improving large language model (LLM) reasoning by allocating additional compute during inference, e.g., via multi-sample generation and verifier-based reranking. Existing TTC scaling strategies and reasoning scorers remain fragmented, evaluated under inconsistent protocols, and are rarely analyzed through the lens of quality-cost trade-offs. We introduce ThinkBooster, a unified framework for seamless test-time compute scaling of LLM reasoning, which consists of (i) a modular Python library implementing state-of-the-art TTC scaling strategy and scorer families, (ii) a benchmark that jointly evaluates performance and computational efficiency, and (iii) a deployable OpenAI-compatible proxy service that enables drop-in integration of adaptive reasoning into real-world applications. We further provide a demo visual debugger for inspecting the reasoning trajectories, intermediate selection decisions, and alternative reasoning paths. Empirical results on mathematical and coding tasks reveal the performance-compute trade-offs of TTC scaling strategies and scoring methods and demonstrate that ThinkBooster provides practical gains in real-world tasks. The code is available online under an MIT license.

Many modern applications of deep learning involve training a neural network via a one-step prediction loss (e.g., $L^2$ regression, cross-entropy), but deploy the network by rolling out along its own predictions. Key examples include autoregressive language modeling, flow-based generative modeling, and robot policy learning. It is well-documented that these settings induce a phenomenon we call test-time feedback (TTF): the mismatch between the training/validation loss and downstream metrics of interest, such as task success rate and generation quality, which grows with task length. While data curation, architecture, and objective design have been proposed to combat train-test shift in TTF settings, this paper proposes optimization as a new design axis to mitigate error accumulation. Specifically, we introduce a new optimization paradigm called double-preconditioning (DoPr) uniquely tailored to the challenges of TTF. DoPr combines gradient-wise preconditioning, as in Adam and Muon, with activation-wise preconditioning (AP), such as in KFAC. We show that the addition of AP yields a drop-in intervention for increasing downstream model performance across a range of TTF settings. Interestingly, these gains in test-time performance do not consistently accompany improvements in validation loss, opening new questions about how to properly evaluate models trained with one-step supervised objectives.

Modern reasoning models can allocate different amounts of test-time computation, such as thinking tokens, model calls, or compute budget, to different tasks. Existing methods generally drive this allocation by predicted difficulty and spend more compute where it is expected to raise accuracy. This implicitly assumes that all failures cost the same, since an accuracy objective weights every task equally. However, such an assumption does not hold in deployment: A typo in a log message and a migration that corrupts a production database both count as one benchmark failure, but their real-world costs are fundamentally different. To fill this gap, we propose consequence-aware test-time compute allocation. Instead of routing compute only by predicted difficulty, we use a lightweight predictor to estimate from the issue text how costly a task would be if solved incorrectly. The scheduler then routes higher-consequence tasks to larger compute tiers or higher thinking budgets under the same total budget. We conduct main experiments on SWE-bench Lite and evaluate cross-dataset behavior on Multi-SWE-bench mini, covering 700 software-engineering tasks in total. Our results reveal that consequence and difficulty are approximately orthogonal under various annotations, and that current thinking models do not allocate compute sufficiently according to consequence. Moreover, our issue-only predictor never misclassifies a high-consequence task as low-consequence across the 300 SWE-bench tasks. Under matched compute budgets, our consequence-aware scheduler reduces cost-weighted loss by 22% to 33% relative to difficulty-aware routing; in particular, the priority-aware variant, which routes by per-task cost scaled by the marginal-utility signal, crosses 30%, and its deployable predictor-driven version retains over 90% of the oracle gain.

The VRR-QA challenge evaluates visual relational reasoning in videos, where answers often depend on implicit spatial relations, event boundaries, target identity, and dialogue context rather than a single salient frame. We present a test-time reasoning pipeline built around a strong GPT-5.5 video QA solver and a set of question-aware evidence ledgers. The initial solver answers each question from a uniform video representation, while routed ledgers are prompted to make the required targets, count units, reference frames, and temporal or spatial scope explicit for counting, spatial, endpoint, viewpoint, and dialogue reasoning. External tools such as open-vocabulary detection, depth cues, pair crops, ASR, and scene-graph ledgers are used only as evidence sources. A conservative gate keeps the current answer unless independent evidence uniquely supports a different option. The final evidence-gated pipeline achieves 92.95% overall accuracy and 93.79% macro accuracy on the challenge test split.

Existing research largely reduces cultural intelligence in LLMs to a knowledge-level problem, overlooking whether models can effectively utilize their acquired knowledge in realistic scenarios. To bridge this gap, we introduce CultureForest, a benchmark for \textit{Cultural Norm Grounded Reasoning}. Each question is grounded in a small set of atomic norms, enabling verifiable and attributable evaluation. CultureForest comprises 5,378 examples across 8 domains and 53 countries/regions, and supports a progressive evaluation from multiple-choice to open-ended generation. Extensive experiments reveal that even top-tier models degrade substantially in open-ended settings, accompanied by pronounced cross-region disparities. Through targeted analysis, we uncover several consistent patterns: (1) test-time reasoning yields limited gains and may exacerbate inequity; (2) models exhibit highly shared regional preference structures; (3) model responses are markedly conservative, especially under stricter cultural constraints; and (4) by disentangling cultural knowledge acquisition from cultural reasoning, we show that while LLMs possess substantial cultural knowledge, their performance is further bottlenecked by its effective use. These findings point to a necessary shift from knowledge-centric evaluation toward measuring knowledge-grounded reasoning.

VRR-QA evaluates whether video-language systems can infer spatial, temporal, viewpoint, depth, and visibility relations that are not always resolved by a single frame. We present an inference-only system built around adaptive test-time computation. The system first answers each question with a direct video-language model pass, then uses multiple lightweight views to find unstable questions. Only these difficult questions are routed to a high-budget dense evidence module that constructs timestamped frame observations, relation-specific probes, candidate verification, and conservative temporal aggregation. This design separates two problems that are often confused in video question answering: finding plausible alternative answers and deciding when a current answer should actually be changed. On the test split, the final system obtains 90.07 average accuracy and 87.81 macro average accuracy. The report focuses on the final test system and the implementation settings required to reproduce the adaptive dense verifier.

Agentic search requires language model agents to explore many sources and answer complex information-seeking questions. Scaling test-time compute is a promising way to improve these agents, but current approaches can fail, because correct answers are often sparse and score-based selection depends on model calibration. We propose FineVerify, a fine-grained self-verification framework that decomposes each question into checkable sub-questions, verifies sampled candidates against each sub-question, and selects the candidate with the highest aggregated score. This per-check structure turns selection into simpler local judgments and produces scores under the same explicit criteria. Across four agentic search benchmarks and two models, FineVerify consistently outperforms standard scaling baselines. With only four sampled trajectories, it improves GPT-5-mini by 8.2 accuracy points and Gemini-3-flash by 5.6% on average. With 12 samples, FineVerify enables GPT-5-mini to surpass frontier GPT-5 on BrowseComp-Plus. Beyond accuracy, FineVerify produces interpretable verification traces that help audit benchmark errors, suggesting broader applications for inspecting agentic search systems. Code and data are available at https://github.com/XuZhao0/fineverify

Test-time compute (TTC) strategies have emerged as a lightweight approach to boost reasoning in large language models (LLMs). However, their application and benefits for vision-language models (VLMs) remain underexplored. We present a systematic study of TTC across seven VLMs and six benchmarks, specifically analyzing feature-based scoring and majority voting methods. We find that feature heuristics fail and voting yields only modest gains in single-model settings. We theoretically show that this limitation stems from a lack of prediction diversity: when outputs are highly correlated, voting provides little benefit. In contrast, multi-model ensembles offer richer diversity, yet standard majority voting fails to account for varying model capabilities. To address this, we propose Entropy-based TTC (ETTC), which selects the most confident prediction based on predictive entropy. Our method reduces to majority voting in the single-model case, but in model ensembles, it leverages confidence disparities to prioritize stronger models. We prove that ETTC outperforms majority voting under mild assumptions and empirically demonstrate that it consistently surpasses both voting and the best individual model. Crucially, our results show that smaller models can synergistically enhance larger ones, unlocking ensembling gains not achievable with standard strategies.

Test-time finetuning (TTFT) is a rapidly evolving paradigm that adapts a language model to each prompt by retrieving related sequences, updating the model on them, and then evaluating the prompt. However, TTFT is only practical if it is fast: selection and finetuning both happen per query, making each a direct bottleneck. Existing methods trade speed for quality: fast retrieval is often redundant, while stronger diversity-aware selection adds prohibitive per-query cost. We introduce HullFT, a geometric approach to TTFT that addresses both bottlenecks. Given a query, HullFT first represents the query embedding as a sparse convex combination of few training sequences, using efficient projection-free Frank-Wolfe optimization. This yields a support set that is inherently relevant and diverse. We then convert the fractional convex weights into an exact integer multiset for finetuning through a geometric integerization procedure. The resulting multiplicities naturally create repeated examples, which we exploit with Gradient Reuse to amortize forward-backward computation across repeated finetuning steps. Our experiments show that HullFT improves the quality-efficiency tradeoff over current state-of-the-art TTFT methods, achieving lower bits-per-byte at substantially lower total runtime.

When machine learning systems under-perform for particular subgroups, affected users typically have no way to correct these disparities without relying on platform-level fixes. Existing approaches to algorithmic fairness rely on provider-centric approaches to correct these failures, leaving users with no external lever when faced with harm. Recent work in Algorithmic Collective Action shows that coordinated users can steer an algorithmic system toward a collective goal, but the existing mechanisms require the provider to retrain on the collective's modified data which users may not have control over. We propose Test-Time Collective Action (TTCA), a framework through which a group of users who share query access to the platform, can correct disparities affecting under-served subgroup without participating in the platform's training loop. We implement this through a proxy-based mechanism where the collective pools query access to a black-box API to extract a proxy of the platform, then optimizes a per-class universal perturbation against the proxy. Each member applies this perturbation to their own inputs at submission time, requiring no cooperation from the platform. We empirically evaluate the mechanism on CIFAR-10, CIFAR-100, and FairFace, showing that modestly-sized collectives close most of the subgroup accuracy gap, transfer across architectures (a small proxy can attack a larger platform), and improve worst-group accuracy, equal-opportunity gap, and disparate impact. A query-budget analysis comparing a per-user black-box attack baseline shows that pooling is cheaper than each subgroup member attacking alone. Test-time collective action thus offers corrective intervention to users when platform-side remediation is unavailable or delayed.

Repeated sampling with a verifier is the standard way to allocate test-time compute for code generation, with pass@$K$ as the canonical metric. Yet the standard policy class draws $K$ independent samples from a single answer distribution, so attempts often collapse onto near-duplicate reasoning paths and waste the budget on redundant rollouts. This failure is costly in competitive programming, where many problems admit multiple distinct algorithmic strategies and pass@$K$ requires only one correct attempt. We propose Coordinated Pass@$K$ Policy Optimization (CPPO), which turns pass@$K$ generation into joint exploration over strategies: a planner emits a tuple of $K{=}4$ alternative high-level methods, and a shared solver attempts one solution per method. CPPO trains this joint policy with a multiplicative planner reward, $R_{\mathrm{plan}} = J_ψ\cdot R_{\mathrm{out}}$, assigning credit only to valid strategy tuples that lead to verifier-confirmed pass@$K$ success. Across APPS, CodeContests, and LiveCodeBench-v6, CPPO improves pass@$4$ over direct sampling, planning baselines, planner-only SFT, and pass@$K$-oriented RL under the same $K{=}4$ solver-attempt budget, with statistically significant gains on six of nine model--benchmark cells. The largest single gain is $+0.16$ on Qwen3.5-9B LiveCodeBench-v6 over the strongest baseline, PKPO ($0.588 \rightarrow 0.748$; paired bootstrap, $p < 0.05$).

Aligning large language models (LLMs) with diverse and multifaceted user preferences is a fundamental challenge in personalized AI systems. Existing multi-objective alignment methods either rely on costly training or require pre-trained reward models for each preference, making it difficult for them to adapt to evolving preferences. Prompt-based personalization offers a training-free alternative, but prompting alone often provides limited steerability, as LLMs may overemphasize or overlook certain preferences and fail to give users reliable control over the relative importance of different objectives when conflicts arise, leading to suboptimal alignment. In this paper, we introduce MATO, a training-free framework for Multi-objective personalized Alignment with Test-time Optimization. MATO formulates personalization as a test-time optimization problem that steers the relative importance of multiple objectives through controllable weights during decoding, without modifying model parameters or requiring external reward models. Specifically, a reward discovery module recovers preference rewards directly from the backbone LLM for diverse objectives specified in natural language, while a weight optimization module dynamically adjusts objective weights based on the user's initial preferences and the partially generated response to balance competing objectives during generation. The resulting rewards and weights jointly guide an online optimization procedure over the token distribution, enabling better alignment with the target objectives. Extensive experiments across multiple datasets and backbone LLMs show that MATO consistently outperforms strong baselines, achieving Pareto-improving multi-objective alignment and stronger steerability. These results highlight test-time optimization as a promising direction for scalable, controllable, and model-agnostic personalized alignment.

Scaling test-time compute has proven highly effective for language models, yet this opportunity remains largely unexplored for industrial Click-Through Rate (CTR) prediction. CTR models suffer from a fundamental asymmetry: feature combinations well-represented in training yield confident predictions, while sparsely observed ones produce unreliable outputs. Existing training-phase solutions such as adaptive gating learn a fixed selection function subject to the same sparsity, offering no per-instance recourse at deployment.We propose UTTSI (Uncertainty-Triggered Test-Time Selective Inference), a training-free model-agnostic framework that scales inference depth proportionally to per-instance uncertainty. A dual-signal estimator combining model logit confidence with a data-level frequency prior distinguishes epistemic uncertainty from aleatoric ambiguity. Every instance undergoes adaptive feature filtering to remove unreliable embeddings; uncertain instances additionally receive stochastic feature-path explorations whose predictions are aggregated via consistency-weighted ensembling. Confident instances bypass exploration entirely, keeping average overhead at approximately $2.8\times$ base model cost with worst-case latency unchanged.Experiments on four datasets with three backbone architectures demonstrate consistent, statistically significant gains over all training-phase baselines. A seven-day online A/B test further confirms a 5.3% relative CTR gain ($p < 0.01$), establishing selective test-time compute allocation as a practical complement to training-phase advances for CTR prediction.

With the continuous advancement of Large Language Models (LLMs), intelligent agents are becoming increasingly vital. However, these agents often fail in environments governed by implicit rules--hidden constraints that cannot be observed directly and must be inferred through interaction. This causes agents to fall into repetitive trial-and-error loops, ultimately leading to task failure. To address this challenge, we propose Test-Time Exploration (TTExplore), a framework where a thinker component analyzes interaction history to infer these implicit rules and guide an actor. Effective exploration in this setting critically depends on the reasoning ability of the thinker. However, evaluating deep reasoning trajectories is inherently unstable and difficult, which poses a major obstacle to effective training. To overcome this issue, we introduce a novel and stable reinforcement learning pipeline. The core idea is to use accurate task-level scores as indirect rewards to bypass the difficulty of evaluating intermediate reasoning, and to retain only a single thinking node per trajectory to alleviate reward sparsity. Using this pipeline, we train a specialized 7B model, Exp-Thinker. Experiments on five text-based embodied tasks show that TTExplore equipped with Exp-Thinker improves baseline agent performance by an average of $14$-$19$ points, demonstrating the effectiveness of explicitly reasoning about implicit rules.

Recently, Reinforcement Learning with Verifiable Rewards (RLVR) and Test-Time Scaling (TTS) have advanced LLM code generation through executable verification. Yet Ground-Truth Unit Tests (GT UTs) remain a bottleneck: SOTA RLVR methods require them for costly training, while existing TTS methods lose competitiveness without them. This motivates GT-free TTS, where existing methods directly use self-generated UTs to refine and select code candidates. Yet such UTs are often noisy or spuriously coupled with wrong code, and UT quality in turn cannot be validated without reliable code. The key challenge is therefore to jointly improve both. To this end, we present CoSPlay, a GT-free, training-free framework that jointly improves codes and UTs through cooperative self-play. It first explores diverse solution ideas and identifies their potential failure modes to produce discriminative UT ideas. It then uses bidirectional pass-count signals from the Code-UT execution matrix to iteratively prune or fix weak codes and refresh or replace unreliable UTs, letting the two pools co-evolve. Finally, when multiple codes remain tied at the highest pass count, it picks the final code from the largest output-consensus cluster, since correct codes agree on the same inputs while wrong codes diverge. Experiments on four challenging benchmarks show that CoSPlay on Qwen2.5-7B-Instruct improves average BoN from 22.1% to 33.2% and UT accuracy from 14.6% to 78.3%, matching or surpassing the RLVR model CURE-7B. When applied to CURE-7B, it further improves BoN by 5.7%. CoSPlay also generalizes across diverse backbones and outperforms GT-free TTS baselines under comparable token budgets, with continued gains as the budget scales up. These results suggest a scalable inference strategy for competitive code generation without any GT data.

Language models must now generalize out of the box to novel environments and work inside inference-scaling search procedures, such as AlphaEvolve, that select rollouts with a variety of task-specific reward functions. Unfortunately, the standard paradigm of LLM post-training optimizes a pre-specified scalar reward, often leading current LLMs to produce low-entropy response distributions and thus to struggle at displaying the diversity that inference-time search will require. We propose Vector Policy Optimization (VPO), an RL algorithm that explicitly trains policies to anticipate diverse downstream reward functions and to produce diverse solutions. VPO exploits that rewards are often vector-valued in practice, like per-test-case correctness in code generation or, say, multiple different user personas or reward models. VPO is essentially a drop-in replacement for the GRPO advantage estimator, but it trains the LLM to output a set of solutions where individual solutions specialize to different trade-offs in the vector reward space. Across four tasks, VPO matches or beats the strongest scalar RL baselines on test-time search (e.g. pass@k and best@k), with the gap widening as the search budget grows. For evolutionary search, VPO models unlock problems that GRPO models cannot solve at all. As test-time search becomes more standardized, optimizing for diversity may need to become the default post-training objective.