Test Time

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.

Test-Time Scaling (TTS) enhances the reasoning capabilities of large language models by allocating additional inference compute to explore the solution space. However, existing parallel TTS methods typically keep branches isolated during search: intermediate discoveries remain branch-private and cannot guide other branches in time. This information isolation causes substantial redundant exploration, as branches repeatedly rediscover information already found elsewhere and require more search steps to collect complete decision information needed to reach correct answers. To bridge this gap, we propose \textbf{Collaborative Parallel Thinking (CPT)}, a training-free inference framework that enables search-time information sharing across parallel branches. CPT extracts compact intermediate information from ongoing branches, maintains a deduplicated query-level information pool, and broadcasts pool entries through the input context, allowing each branch in subsequent search steps to reuse discoveries made by other branches rather than rediscover the same information. Empirically, experiments on HMMT and AIME benchmarks show that CPT establishes a stronger accuracy--latency Pareto frontier than strong baselines across rollout budgets and model scales, highlighting search-time collaboration as an effective direction for efficient parallel TTS.

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.

Test-time scaling improves language model reasoning by spending additional compute to explore multiple solution trajectories. The key challenge is to maximize accuracy while minimizing the total number of generated tokens during reasoning. Recent PRM-guided methods score intermediate prefixes to steer this search, but most are frontier-only: they keep only the current active prefixes and irreversibly prune or resample away the rest using noisy PRM scores. This can cause premature commitment, diversity collapse, and the loss of prefixes that still admit correct continuations. We introduce stochastic backtracking over a persistent pool of historical prefixes, allowing test-time compute to revisit previously generated states instead of only expanding the current frontier. To make this efficient, we propose two complementary mechanisms. Subpool Selection strengthens greedy PRM-guided search by applying Top-N selection within random subpools, giving historical prefixes a chance to bypass over-scored frontier candidates. Power Backtrack Sequential Monte Carlo extends SMC-style resampling to the persistent pool using powered PRM scores and mixture-corrected weights. Across mathematical reasoning benchmarks and model scales, our methods consistently achieve higher accuracy per token count, and the same level of accuracy using only a fraction of the token count in comparison to strong PRM-guided baselines, demonstrating that persistent-pool stochastic backtracking provides a simple and effective way to improve the accuracy-token trade-off in test-time scaling.

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.

Test-time scaling has emerged as a promising approach for improving code generation by exploring large solution spaces at inference time. However, existing methods often rely on public test cases that are unavailable in practice, or require extensive LLM inference for candidate selection, leading to significant token consumption and time overhead. We present DiffCodeGen, a novel test-time scaling method for code generation based on coverage-guided differential analysis. DiffCodeGen generates diverse code candidates using various sampling and prompting strategies, then applies coverage-guided fuzzing to synthesize inputs without requiring any existing tests or large language models. By executing all candidates on these inputs, DiffCodeGen captures their dynamic behavior and clusters candidates based on behavioral similarity. DiffCodeGen selects the medoid of the largest cluster as the final output. Unlike prior test-time scaling methods that invoke additional LLM inference for candidate selection, DiffCodeGen performs selection without any extra model calls, incurring little to no additional token consumption. DiffCodeGen is fully asynchronous, naturally suited to the current trend of agentic coding, and is thus efficient and highly scalable. We evaluate DiffCodeGen across 4 large language models, demonstrating consistent improvements over baselines. Compared to state-of-the-art test-time scaling methods, DiffCodeGen achieves competitive or superior performance while using only a fraction of time and tokens. DiffCodeGen is model-agnostic and can be combined with reasoning models to further boost performance.

Test-time adaptation (TTA) enables a pre-trained model to adapt online to an unlabeled test stream under distribution shift. While most TTA research focuses on the adaptation objective, practical streams also depend critically on the memory used to select which test samples drive adaptation. Existing memory mechanisms are usually evaluated as components of specific TTA algorithms, making it difficult to isolate which memory design choices matter and when they matter. In this work, we provide a systematic benchmark that decouples memory from the adaptation algorithm and evaluates memory policies under unified conditions across i.i.d., non-i.i.d., continual, and practical test streams. Our study shows that effective memory management requires more than retaining recent or class-balanced samples. In particular, intra-class diversity is a key factor for avoiding redundant buffers and maintaining representative adaptation signals under temporally correlated and label-skewed streams. Motivated by this finding, we introduce Guided Observational Test-Time Adaptation (GOTTA), a family of diversity-aware memory policies that combine class-balanced allocation with feature-space diversity. GOTTA memories act as drop-in replacements for existing buffers and can be paired with different TTA objectives. Across corruption benchmarks and video-stream settings, diversity-aware memory improves adaptation most clearly under constrained memory budgets and challenging non-i.i.d. streams, while remaining competitive as memory capacity increases. These results highlight memory management as a first-class component of robust test-time adaptation and identify diversity as a central principle for practical TTA.

Large language models (LLMs) often expose useful signals of self-monitoring: before solving a problem, they can estimate whether they are likely to succeed, and after solving it, they can judge whether their answer is likely to be correct. However, these signals are typically measured or elicited in isolation, rather than used to control inference. In this work, we ask whether LLMs possess latent metacognitive ability that can be turned into effective test-time control. Inspired by the Nelson--Narens theory from cognitive psychology, we propose a metacognitive harness that separates monitoring from reasoning. For each problem, the model first reports a pre-solve feeling-of-knowing (FOK) signal; after each solve attempt, it reports a post-solve judgment-of-learning (JOL) signal. Rather than treating these signals as passive confidence estimates, the harness turns them into an explicit control interface for reasoning: it decides when to trust the current solution, when to retry with compact metacognitive feedback, and when to pass multiple attempts to a final aggregator. Across text, code, and multimodal reasoning benchmarks, our harness substantially improves a fixed Claude Sonnet-4.6 base model without parameter updates or benchmark-specific fine-tuning. On the evaluated public benchmark snapshots, it raises pooled accuracy from 48.3 to 56.9 and exceeds the strongest listed leaderboard entries on the three primary evaluation settings: HLE-Verified, LiveCodeBench v6, and R-Bench-V. These results suggest that strong LLMs may already possess useful metacognitive ability, but require an explicit control harness to act on it during reasoning.

Test-time scaling (TTS) methods have proven highly effective for LLMs, yet their application to vision-language models (VLMs) remains relatively underexplored. Existing VLM TTS methods largely require open-weight model access or expensive repeated sampling, and are evaluated primarily on multimodal mathematical and scientific reasoning benchmarks rather than general visual understanding tasks. In this paper, we propose Test-Time Hinting, a method that improves VLM performance via a single VLM call and requiring only black-box API access, which makes it broadly applicable to frontier closed-weight models. Our method is motivated by the observation that VLM errors tend to cluster around recurring failure patterns. We therefore train a lightweight hint generator model to predict, for a given test input, which "hint" should be prepended to the prompt, providing targeted contextual or procedural guidance that steers the VLM away from its characteristic failure modes. We show that Test-Time Hinting improves the accuracy of multiple closed-weight VLMs on natural-image VQA benchmarks and that these gains generalize to unseen benchmarks and VLMs without retraining the hint generator.

Test-time compute is widely believed to benefit only large reasoning models, leaving small models with nothing to gain. We argue the opposite for dense retrieval, since modern small embedding models are distilled or adapted from large language model backbones and can inherit their latent test-time-compute potential. We ask how much retrieval quality a frozen embedding model gains at inference alone, with no auxiliary model and no parameters trained at deployment. An agentic loop in which a large language model writes programs over a frozen encoder API explores 144 candidates and yields twelve Pareto-optimal programs that trade inference compute for quality across cost ratios from $c{=}1.2$ to $14.7$, every one improving nDCG@10 on all 14 discovery tasks. The programs use no trainable parameters and recover classical retrieval primitives, among them reciprocal rank fusion, the Fisher linear discriminant, Rocchio pseudo-relevance feedback, and sentence-level MaxSim. Applied unmodified to nineteen held-out tasks and three unseen encoder families, a single fixed program improves the majority of tasks, with a positive median $Δ$nDCG@10 and a 54 to 57% win-rate at $c{\ge}4$, and the gains are largest on encoder families never seen during discovery. A matched-budget learned projection head trained on the same tasks does not transfer this way, improving in-domain retrieval by $+0.20$ to $+0.25$ nDCG@10 yet falling below baseline on every held-out encoder. Small embedding models therefore inherit usable test-time-compute potential, and a frozen encoder converts inference compute into retrieval gains that transfer to new corpora and encoders with no per-domain labels.

Automated scientific discovery using large language models relies on identifying genuinely novel solutions. Standard reinforcement learning penalizes high-variance mutations, which leads the policy to prioritize familiar patterns. As a result, the maximum reward plateaus even as the average reward increases. Overcoming this limitation requires a signal that distinguishes unexplored regions from intrinsically difficult problems. This necessitates measuring disagreement across independently adapted weight hypotheses rather than relying on a single network's confidence. UG-TTT addresses this challenge by maintaining a small ensemble of low-rank adapters over a frozen base model. The per-token disagreement, quantified as the mutual information between ensemble predictions and weight hypotheses, isolates epistemic uncertainty and identifies positions where insufficient coverage leads to adapter divergence rather than intrinsic problem difficulty. This measure is incorporated as an exploration bonus into the policy gradient, directing the policy toward positions where persistent adapter disagreement signals low training coverage, the same frontier where genuine discovery is possible. A nuclear norm regularizer ensures the adapters remain distinct from one another, thereby preserving the exploration signal throughout training. Across four scientific discovery benchmarks, UG-TTT increases the maximum reward on three tasks, maintains substantially higher solution diversity, and an ablation study confirms that the regularizer is essential for sustaining this behavior.

Large language models are increasingly deployed with test-time strategies: sample $N$ responses, score them with a reward model or verifier, and return the best. This deployment rule exposes a mismatch in post-training: standard objectives optimize the mean reward of a single response, whereas best-of-$N$ performance is governed by the upper tail of the reward distribution. Recent test-time-aware objectives partly address this mismatch, but typically assume that training can use the same per-prompt rollout budget as deployment, which is impractical when post-training must cover many prompts while deployment can allocate much larger per-prompt test-time compute. We study this budget-mismatch regime, where only $m\ll N$ per-prompt rollouts are available during training but the target objective is best-of-$N$ deployment. Under structural assumptions on the reward tails, we show that the policy gradient of the best-of-$N$ objective can be approximated from a much smaller rollout group by extrapolating upper-tail statistics. This yields a family of Tail-Extrapolated estimators for best-of-$N$-oriented post-training: a simple direct estimator, Tail-Extrapolated Advantage (TEA), and a fixed-order debiased Prefix-TEA estimator based on moment cancellation. Experiments on instruction-following tasks show that TEA and Prefix-TEA improve best-of-$N$ performance across different language models, reward models and datasets under various training and test-time budget settings.

Experience-driven self-evolving agents aim to overcome the static nature of large language models by distilling reusable experience from past interactions, thus enabling adaptation to novel tasks at deployment time. This process places substantial demands on the foundation model's capacities for abstraction, generalization, and in-context learning. However, most existing studies focus primarily on system-level design choices, such as how experience is represented and managed, neglecting the inherent capabilities of the underlying model. While some recent works have started to optimize the experience utilization stage via reinforcement learning, they still fail to treat self-evolution as a unified process to be jointly optimized. To this end, we propose Evolving-RL, an efficient algorithmic framework that jointly improves the experience extraction and utilization capabilities required for self-evolution. Specifically, we center the learning process on experience extraction and evaluation, using the two supervisory signals derived from evaluation to optimize the extractor and solver separately and thus enable their coordinated co-evolution. Experiments on ALFWorld and Mind2Web show that Evolving-RL effectively enhances LLMs' ability to extract and reuse experience, leading to strong performance gains on out-of-distribution tasks (up to 98.7% relative improvement over the GRPO baseline on ALFWorld unseen tasks and 35.8% on Mind2Web), and these gains are fully unlocked only through the coordinated co-evolution of experience extraction and utilization. Furthermore, Evolving-RL inherently functions as an experience-augmented RL algorithm. By internalizing reusable experience patterns directly into model parameters, it achieves remarkable performance gains over standard baselines on both seen and unseen tasks, even in the absence of test-time experience accumulation.

Memory consolidation, the process by which transient experiences are transformed into stable, structured representations, is a foundational organizing principle in the human brain, yet it remains largely unexplored as a design principle for modern sequence models. In this work, we leverage established neuroscientific theories of memory consolidation and cross-frequency coupling to propose the Hierarchical Memory Module (HMM), a neural memory architecture composed of two functionally distinct sub-modules that operate at different update frequencies. Inspired by the transformation hypothesis, the low-frequency sub-module produces high-level representations that capture abstract, gist-level knowledge, while the high-frequency sub-module produces fine-grained representations that preserve richer episodic detail. The final memory output is dynamically reconstructed as a context-dependent combination of both representations, analogous to the reconstructive nature of human memory retrieval. We integrate HMM into a Transformer-based language decoder to form Mela, a family of memory-augmented language models that perform online memory consolidation at test time. To further exploit the multi-granularity memory representations produced by HMM, we introduce MemStack, a method that distributes different levels of memory features across the early layers of the decoder without introducing additional tokens. Experiments on language modeling demonstrate that Mela outperforms Transformer baselines across all the model sizes. Moreover, with the pretrained context length fixed at 4K, Mela maintains performance on significantly longer contexts, whereas Transformer baselines degrade rapidly beyond their training length. Extensive ablation studies validate the contribution of each component and provide guidance for practical configuration.

Test-time scaling has become an effective paradigm for improving the reasoning ability of large language models by allocating additional computation during inference. Recent structured approaches have further advanced this paradigm by organizing inference across multiple trajectories, refinement rounds, and verification-based feedback. However, existing structured test-time scaling methods either weakly coordinate parallel reasoning trajectories or rely on noisy historical information without explicitly deciding what should be retained and reused, limiting their ability to balance exploration and exploitation. In this work, we propose TMAS, a framework for scaling test-time compute via multi-agent synergy. TMAS organizes inference as a collaborative process among specialized agents, enabling structured information flow across agents, trajectories, and refinement iterations. To support effective cross-trajectory collaboration, TMAS introduces hierarchical memories: the experience bank reuses low-level reliable intermediate conclusions and local feedback, while the guideline bank records previously explored high-level strategies to steer subsequent rollouts away from redundant reasoning patterns. Furthermore, we design a hybrid reward reinforcement learning scheme tailored to TMAS, which jointly preserves basic reasoning capability, enhances experience utilization, and encourages exploration beyond previously attempted solution strategies. Extensive experiments on challenging reasoning benchmarks show that TMAS achieves stronger iterative scaling than existing test-time scaling baselines, with hybrid reward training further improving scaling effectiveness and stability across iterations. Code and data are available at https://github.com/IQuestLab/tmas.

Chain-of-thought (CoT) reasoning improves large language models (LLMs) on difficult tasks, but it also makes inference expensive because every intermediate step must be generated as a discrete token. Latent reasoning reduces visible token generation by propagating continuous states, yet replacing explicit derivations with latent computation can hurt tasks that require symbolic checking. We propose Latent-Then-Explicit Reasoning (LaTER), a two-stage paradigm that first performs bounded exploration in a continuous latent space and then switches to explicit CoT for verification and answer generation. In a training-free instantiation, LaTER projects final-layer hidden states back to the input embedding space, preserves the latent KV cache, and uses entropy and model-native stop-token probes to decide when to switch. We find that strong reasoning models already exhibit structured latent trajectories under this interface. On Qwen3-14B, training-free LaTER reduces total token usage by 16%-32% on several benchmarks while matching or improving accuracy on most of them; for example, it improves AIME 2025 from 70.0% to 73.3% while reducing tokens from 15,730 to 10,661. We further construct Latent-Switch-69K, a supervised corpus that pairs condensed solution intuitions with shortened explicit derivations. Fine-tuning with latent rollout and halting supervision yields additional gains: trained LaTER reaches 80.0% accuracy on AIME 2025, 10.0 points above the standard CoT baseline, while using 33% fewer tokens. Our code, data, and model are available at https://github.com/TioeAre/LaTER.

Beyond training-time optimization, scaling test-time computation has emerged as a key paradigm to extend the reasoning capabilities of Large Language Models (LLMs). However, most existing methods adopt a rigid Planning-before-Trial (PbT) policy, which inefficiently allocates test-time compute by incurring planning overhead even on directly solvable problems. We propose Planning-after-Trial (PaT), an adaptive policy for code generation that invokes a planner only upon verification failure. This adaptive policy naturally enables a heterogeneous model configuration: a cost-efficient model handles generation attempts, while a powerful model is reserved for targeted planning interventions. Empirically, across multiple benchmarks and model families, our approach significantly advances the cost-performance Pareto frontier. Notably, our heterogeneous configuration achieves performance comparable to a large homogeneous model while reducing inference cost by approximately 69%.

This paper studies test-time aggregation, an approach that generates multiple reasoning traces and aggregates them into a final answer. Most existing methods rely on evaluation signals collected from candidate traces in isolation or answer frequencies, while ignoring comparative interactions among candidates. We propose Joint Consistency (JC), formulated as a constrained Ising-type energy minimization problem, where independent evaluation signals act as external fields and pairwise comparisons act as interactions. JC provides a unified framework for test-time aggregation that subsumes existing voting and weighted aggregation methods as special cases. Our construction of the interaction matrix leverages LLM-as-a-judge comparisons, and admits a theoretical interpretation under answer-level homogeneity assumptions. Moreover, we develop an efficient approximation strategy that makes interaction modeling practical for large-scale test-time aggregation. Experiments on math and code reasoning benchmarks show that JC consistently outperforms existing baselines across tasks, judge models, trace budgets, and trace-generation settings.

Black-box optimization in science and engineering often comes with side information: experts, simulators, pretrained predictors, or heuristics can suggest which candidates look promising. This information can accelerate search, but it can also be biased, input-dependent, or misleading. Feedback-aware BO methods typically handle one task at a time, limiting their ability to generalize over multiple sources of feedback. In-context optimizers address cross-task adaptation, but usually assume that optimization history is the only available signal at test time. We study feedback-informed in-context black-box optimization (FICBO), where a pretrained optimizer conditions on both the observed history and cheap auxiliary feedback for the current candidate set. We introduce a structured feedback prior that models how feedback sources vary in their access, relevance, and distortion relative to the true objective, and use it to pretrain a feedback-aware transformer. At test time, the model estimates source reliability in context by comparing observed objective values with auxiliary signals, improving query selection. On synthetic and real-world tasks, FICBO effectively exploits informative feedback while remaining robust to weak or misleading sources, improving over other baselines. Empirical investigations further illustrate how the model perceives test-time sources, offering insights into its interpretability and decision-making process.

Majority voting is one of the few black-box interventions that can improve a fixed stochastic predictor: repeated access can be cheaper than changing a high-capability model. Classical fixed-competence theory makes this intervention look monotone -- more votes help above the majority threshold and hurt below it. We show that this picture is fundamentally incomplete. Under the de Finetti representation for exchangeable repeated correctness, voting is governed by a latent distribution of per-example correctness probabilities. Even simple latent mixtures can generate sharply different voting curves, including nonmonotone behavior and, in an explicit construction, infinitely many trend changes. The full latent law determines the curve, but the curve does not determine the law. The exact object recovered by voting is a signed voting signature: at each binomial variance scale, it records excess latent mass above rather than below the majority threshold. Our main theorem proves that the complete odd-budget curve and this signature are equivalent: the curve increments are signed Hausdorff moments, and the full curve recovers the signature uniquely. This viewpoint explains shape phenomena, branch-symmetric nonidentifiability, realizability, variation, and endpoint rates. It also separates estimation regimes: direct per-example success-probability information targets the full signature, whereas fixed-depth grouped labels reveal only a finite prefix.

Visual Foresight VLA (VF-VLA) has become a prominent architectural choice in the recent VLA due to its impressive performance. Nevertheless, the inherent design of VF-VLA makes it particularly vulnerable to out-of-distribution (OOD) shifts. Because the quality of action directly depends on the accuracy of the predicted future visual information, OOD conditions affect both stages at once. To address this vulnerability, we propose Test-Time Training Visual Foresight VLA ($T^3$VF), a test-time training approach motivated by the observation that the predicted future image and its subsequent observation form a natural supervision pair. To further address the practical challenges that arise from indiscriminate test-time updates, we introduce an adaptive update filtering mechanism. Empirically, $T^3$VF mitigates the OOD vulnerability of VF-VLA at a modest additional inference cost, without requiring any architectural modification or auxiliary modules.

While Test-Time Scaling (TTS) offers a promising direction to enhance video generation without the surging costs of training, current test-time video generation methods based on diffusion models suffer from exorbitant candidate exploration costs and lack temporal guidance. To address these structural bottlenecks, we propose shifting the focus to streaming video generation. We identify that its chunk-level synthesis and few denoising steps are intrinsically suited for TTS, significantly lowering computational overhead while enabling fine-grained temporal control. Driven by this insight, we introduced Stream-T1, a pioneering comprehensive TTS framework exclusively tailored for streaming video generation. Specifically, Stream-T1 is composed of three units: (1) Stream -Scaled Noise Propagation, which actively refines the initial latent noise of the generating chunk using historically proven, high-quality previous chunk noise, effectively establishes temporal dependency and utilizing the historical Gaussian prior to guide the current generation; (2) Stream -Scaled Reward Pruning, which comprehensively evaluates generated candidates to strike an optimal balance between local spatial aesthetics and global temporal coherence by integrating immediate short-term assessments with sliding-window-based long-term evaluations; (3) Stream-Scaled Memory Sinking, which dynamically routes the context evicted from KV-cache into distinct updating pathways guided by the reward feedback, ensuring that previously generated visual information effectively anchors and guides the subsequent video stream. Evaluated on both 5s and 30s comprehensive video benchmarks, Stream-T1 demonstrates profound superiority, significantly improving temporal consistency, motion smoothness, and frame-level visual quality.

TabPFN has recently gained attention as a foundation model for tabular datasets, achieving strong performance by leveraging in-context learning on synthetic data. However, we find that TabPFN is vulnerable to label shift, often overfitting to the majority class in the training dataset. To address this limitation, we propose DistPFN, the first test-time posterior adjustment method designed for tabular foundation models. DistPFN rescales predicted class probabilities by downweighting the influence of the training prior (i.e., the class distribution of the context) and emphasizing the contribution of the model's predicted posterior, without architectural modification or additional training. We further introduce DistPFN-T, which incorporates temperature scaling to adaptively control the adjustment strength based on the discrepancy between prior and posterior. We evaluate our methods on over 250 OpenML datasets, demonstrating substantial improvements for various TabPFN-based models in classification tasks under label shift, while maintaining strong performance in standard settings without label shift. Code is available at this repository: https://github.com/seunghan96/DistPFN.

Task-based dialogue systems assist users in achieving specific goals, such as executing actions or retrieving information, through natural language interactions. Accurate coreference resolution is essential, as it involves identifying object references within the dialogue - a task that becomes increasingly challenging in visually grounded environments characterized by complex scenes and diverse object metadata. However, coreference resolution in task-based dialogue remains limited by poor generalization across domains and heavy reliance on supervised models that often overfit to dataset-specific artifacts. In this work, we propose a unimodal test-time reasoning approach that enables large language models (LLMs) to reason over detailed object metadata and dialogue history to improve coreference resolution. Empirical results on the SIMMC 2.1 dataset demonstrate that LLMs can generate step-by-step reasoning processes that effectively align dialogue context with objects present in the scene. Extensive experiments highlight the models' ability to link conversations and objects accurately. Moreover, we show that test-time reasoning under few-shot settings generalizes effectively to unseen scenarios and novel objects, outperforming encoder-based supervised methods in cross-domain evaluations. These findings underscore the critical role of structured metadata and careful prompt engineering in enhancing the robustness and generalization of task-oriented dialogue systems.

Large Language Models (LLMs) solve many reasoning tasks via chain-of-thought (CoT) prompting, but smaller models (about 7 to 8B parameters) still struggle with multi-step reasoning under tight compute and token budgets. Existing test time reasoning methods such as self consistency (sampling multiple rationales and voting), Tree-of-Thoughts (search over intermediate thoughts), and critique revise loops improve performance, but often at high token cost and without fine-grained step-level control. This project1 aims to address that gap: can Small Language Models (SLMs) reason reliably using the same or fewer tokens? This question is both scientific and practical. Scientifically, it probes whether process supervision and simple test-time controls (such as token budgets and rejection of redundant steps) can substitute for model scale or large sampling counts. Practically, many deployments (on-device, low-latency, or cost-constrained settings) cannot afford huge models or dozens of sampled rationales per query. A method that improves SLM reasoning at fixed cost would therefore be directly useful.

Scaling test-time compute has emerged as a powerful mechanism for enhancing Large Language Model (LLM) performance. However, standard post-training paradigms, Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), optimize the likelihood of individual samples under a base policy, creating a misalignment with test time procedures that rely on aggregated or filtered outputs. In this work, we propose Compute Aligned Training, which aligns training objectives with test-time strategies. By conceptualizing inference strategies as operators on the base policy, we derive new loss functions that maximize performance when said strategies are applied. We instantiate such loss functions for SFT and RL across common test time strategies. Finally, we provide empirical evidence that this training method substantially improves test time scaling over standard training.

While scaling test-time compute can substantially improve model performance, existing approaches either rely on static compute allocation or sample from fixed generation distributions. In this work, we introduce a test-time compute allocation framework that jointly adapts where computation is spent and how generation is performed. Our method begins with a warm-up phase that identifies easy queries and assembles an initial pool of question-response pairs from the test set itself. An adaptive phase then concentrates further computation on unresolved queries while reshaping their generation distributions through evolving in-context demonstrations -- conditioning each generation on successful responses from semantically related queries rather than resampling from a fixed distribution. Experiments across math, coding, and reasoning benchmarks demonstrate that our approach consistently outperforms existing baselines while consuming substantially less inference-time compute.

Test-time scaling improves the reasoning performance of large language models but often results in token-inefficient overthinking, where models continue reasoning beyond what is necessary for a correct answer. Existing dynamic early-exit methods typically rely on single-step confidence signals, which are often unreliable for detecting reasoning convergence in multi-step settings. To mitigate this limitation, we propose TRACE, a training-free framework for efficient test-time scaling that determines when to terminate reasoning based on temporal aggregation of multi-step evidence rather than instantaneous signals. TRACE detects reasoning convergence over time by aggregating two complementary signals across recent reasoning steps: answer consistency, capturing the persistence of predicted answers, and confidence trajectory, modeling the temporal evolution of model confidence. Benefiting from these two factors, TRACE can accurately determine whether the reasoning process has converged, thereby promptly halting inference and effectively avoiding redundant reasoning steps. Extensive experiments on multiple challenging benchmarks show that TRACE reduces reasoning token usage by 25-30% on average while maintaining accuracy within 1-2% of full-length reasoning, consistently outperforming existing dynamic reasoning methods.

The rise of autonomous AI agents suggests that dynamic benchmark environments with built-in feedback on scientifically grounded tasks are needed to evaluate the capabilities of these agents in research work. We introduce Stargazer, a scalable environment for evaluating AI agents on dynamic, iterative physics-grounded model-fitting tasks using inference on radial-velocity (RV) time series data. Stargazer comprises 120 tasks across three difficulty tiers, including 20 real archival cases, covering diverse scenarios ranging from high-SNR single-planet systems to complex multi-planetary configurations requiring involved low-SNR analysis. Our evaluation of eight frontier agents reveals a gap between numerical optimization and adherence to physical constraints: although agents often achieve a good statistical fit, they frequently fail to recover correct physical system parameters, a limitation that persists even when agents are equipped with vanilla skills. Furthermore, increasing test-time compute yields only marginal gains, with excessive token usage often reflecting recursive failure loops rather than meaningful exploration. Stargazer presents an opportunity to train, evaluate, scaffold, and scale strategies on a model-fitting problem of practical research relevance today. Our methodology to design a simulation-driven environment for AI agents presumably generalizes to many other model-fitting problems across scientific domains. Source code and the project website are available at https://github.com/AIPS-UofT/Stargazer and https://aips-uoft.github.io/Stargazer/, respectively.

Tokenization is a key component of autoregressive (AR) generative models, converting raw data into more manageable units for modeling. Commonly, tokens describe local information, such as regions of pixels in images or word pieces in text, and AR generation predicts these tokens in a fixed order. A worthwhile question is whether token structures affect the ability to steer the generation through test-time search, where multiple candidate generations are explored and evaluated by a verifier. Using image generation as our testbed, we hypothesize that recent 1D ordered tokenizers with coarse-to-fine structure can be more amenable to search than classical 2D grid structures. This is rooted in the fact that the intermediate states in coarse-to-fine sequences carry semantic meaning that verifiers can reliably evaluate, enabling effective steering during generation. Through controlled experiments, we find that AR models trained on coarse-to-fine ordered tokens exhibit improved test-time scaling behavior compared to grid-based counterparts. Moreover, we demonstrate that, thanks to the ordered structure, pure test-time search over token sequences (i.e., without training an AR model) can perform training-free text-to-image generation when guided by an image-text verifier. Beyond this, we systematically study how classical search algorithms (best-of-N, beam search, lookahead search) interact with different token structures, as well as the role of different verifiers and AR priors. Our results highlight the impact of token structure on inference-time scalability and provide practical guidance for test-time scaling in AR models.

Test-time scaling has become a powerful way to improve large language models. However, existing methods are best suited to short, bounded outputs that can be directly compared, ranked or refined. Long-horizon coding agents violate this premise: each attempt produces an extended trajectory of actions, observations, errors, and partial progress taken by the agent. In this setting, the main challenge is no longer generating more attempts, but representing prior experience in a form that can be effectively selected from and reused. We propose a test-time scaling framework for agentic coding based on compact representations of rollout trajectories. Our framework converts each rollout into a structured summary that preserves its salient hypotheses, progress, and failure modes while discarding low-signal trace details. This representation enables two complementary forms of inference-time scaling. For parallel scaling, we introduce Recursive Tournament Voting (RTV), which recursively narrows a population of rollout summaries through small-group comparisons. For sequential scaling, we adapt Parallel-Distill-Refine (PDR) to the agentic setting by conditioning new rollouts on summaries distilled from prior attempts. Our method consistently improves the performance of frontier coding agents across SWE-Bench Verified and Terminal-Bench v2.0. For example, by using our method Claude-4.5-Opus improves from 70.9% to 77.6% on SWE-Bench Verified (mini-SWE-agent) and 46.9% to 59.1% on Terminal-Bench v2.0 (Terminus 1). Our results suggest that test-time scaling for long-horizon agents is fundamentally a problem of representation, selection, and reuse.

This beta technical report asks how reusable experience should be represented so that it can function as effective test-time control and as a substrate for iterative evolution. We study this question in 4.590 controlled trials across 45 scientific code-solving scenarios. We find that documentation-oriented Skill packages provide unstable control: their useful signal is sparse, and expanding a compact experience object into a fuller documentation package often fails to help and can degrade the overall average. We further show that representation itself is a first-order factor. A compact Gene representation yields the strongest overall average, remains competitive under substantial structural perturbations, and outperforms matched-budget Skill fragments, while reattaching documentation-oriented material usually weakens rather than improves it. Beyond one-shot control, we show that Gene is also a better carrier for iterative experience accumulation: attached failure history is more effective in Gene than in Skill or freeform text, editable structure matters beyond content alone, and failure information is most useful when distilled into compact warnings rather than naively appended. On CritPt, gene-evolved systems improve over their paired base models from 9.1% to 18.57% and from 17.7% to 27.14%. These results suggest that the core problem in experience reuse is not how to supply more experience, but how to encode experience as a compact, control-oriented, evolution-ready object.

Standard Transformers have a fixed computational depth, limiting their ability to generalize to tasks that require variable-depth reasoning. The usual remedy, Chain-of-Thought (CoT), spends tokens to reason, inflating the key--value cache and making latency grow with the step count, so memory becomes the limiting cost when reasoning is served over large query batches. We study a depth-recurrent Transformer that decouples computational depth from parameter count by iterating a shared-weight block, so that each added reasoning step costs flat memory and linear latency, with no token generation. Three ingredients keep the recurrence stable for 20+ thinking steps: a silent thinking objective that supervises only the final output, LayerScale initialization, and an identity-biased gate that opens a gradient highway across steps. We characterize it on three compositional domains with decreasing structural bias: graph reachability (adjacency masking), nested boolean logic (relative positioning), and unstructured relational text (no positional cue). We find a \emph{computational frontier}: accuracy climbs once the thinking-step count meets the task's complexity, reaching near-perfect performance on the two structured tasks and a lower plateau on unstructured text. How it climbs depends on the structural bias---abruptly from chance on the graph task, gradually on the other two. Depth recurrence extrapolates beyond the training range: it succeeds on the graph task where fixed-depth models barely extrapolate, and on the two sequence tasks comes within two points of fixed-depth Transformers that use $4$--$6.4\times$ more parameters. On the graph task, whose adjacency mask makes propagation depth verifiable, intermediate per-step supervision---a standard recipe for deep iterative models---consistently \emph{harms} this extrapolation. We release the code for reproducibility.

Test-time guidance is a widely used mechanism for steering pretrained diffusion models toward outcomes specified by a reward function. Existing approaches, however, focus on maximizing reward rather than sampling from the true Bayesian posterior, leading to miscalibrated inference. In this work, we show that common test-time guidance methods do not recover the correct posterior distribution and identify the structural approximations responsible for this failure. We then propose consistent alternative estimators that enable calibrated sampling from the Bayesian posterior. We significantly outperform previous methods on a set of Bayesian inference tasks, and set a new state-of-the-art PSNR in black hole image reconstruction.

Test-time training (TTT) adapts large language models (LLMs) during inference using only unlabeled test inputs. Existing methods, however, face two major bottlenecks on hard reasoning tasks: (1) \emph{lack of learnable samples}, as self-generated pseudo-labels on difficult questions are often noisy and yield unstable rewards; and (2) \emph{inefficient exploration}, as performance gains depend on repeatedly sampling many rollouts without explicit diagnosis of why previous attempts fail. We propose \textbf{TTSR} (\textbf{T}est-\textbf{T}ime \textbf{S}elf-\textbf{R}eflection), a self-evolving framework based on a \emph{reflect-then-synthesize} paradigm. A single pretrained model alternates between a \textit{Student} role and a \textit{Teacher} role: the Student solves test questions and updates, while the Teacher analyzes failed trajectories and synthesizes targeted variant questions closer to the Student's capability frontier. TTSR further maintains a cross-iteration \textit{weakness memory} and compiles persistent weaknesses into a lightweight \textit{strategy note} prepended to subsequent Student inputs, so diagnostic knowledge can guide exploration and gradually fade as weaknesses are resolved. Experiments on challenging mathematical reasoning benchmarks show consistent test-time improvements, strong cross-backbone generalization, and transfer to general-domain reasoning tasks.

This paper proposes FreeFuse, a training-free framework for multi-subject text-to-image generation through automatic fusion of multiple subject LoRAs. In contrast to prior studies that focus on retraining LoRAs to alleviate feature conflicts, our analysis shows that spatially routing LoRA residuals to their intended semantic regions provides an effective mechanism for suppressing direct cross-region LoRA interference while preserving the base model's global contextual reasoning. Accordingly, we implement Adaptive Token-Level Routing during the inference phase. However, obtaining reliable routing regions remains challenging. Existing methods that rely on text-image latent association, such as raw cross-attention or concept-level similarity matching, often suffer from sparse activations, hole artifacts, and unstable localization when handling visually similar subjects, leading to incomplete or ambiguous subject masks. To address these issues, we introduce FreeFuseAttn, a mechanism that exploits the flow matching model's intrinsic semantic alignment to dynamically match subject-specific tokens to their corresponding spatial regions at early denoising timesteps, thereby bypassing the need for external segmentors. FreeFuse distinguishes itself through high practicality: it necessitates no additional training, model modifications, or user-defined spatial constraints. Users need only provide subject activation words to achieve seamless integration into standard workflows. Extensive experiments validate that FreeFuse outperforms existing approaches in both identity preservation and compositional fidelity. The project page is at: https://future-item.github.io/FreeFuse/.

Reinforcement learning (RL) promises to expand the capabilities of language models, but it is unclear if current RL techniques promote the discovery of novel behaviors, or simply sharpen those already present in the base model. In this paper, we investigate the value of deliberate exploration -- explicitly incentivizing the model to discover novel and diverse behaviors -- and aim to understand how the knowledge in pre-trained models can guide this search. Our main finding is that exploration with a simple, principled, representation-based bonus derived from the pre-trained language model's hidden states significantly improves diversity and pass@k rates -- both for post-training, and in a novel inference-time scaling setting we introduce. For inference-time, exploration with representation-based diversity improves efficiency, consistently improving pass@k rates across a variety of models and reasoning tasks. For example, for Qwen-2.5-14b-Instruct we obtain over 50% improvement in verifier efficiency on almost all tasks. For post-training, we show that integrating this exploration strategy into an RL pipeline improves reasoning performance over that of the initial model and over standard RL post-training. For example, on AIME 2024, our post-trained Qwen-2.5-7b-Instruct's pass@80 matches the pass@256 of GRPO on the same model, demonstrating a 3x improvement in test-time sample efficiency. Overall, our findings suggest that deliberate exploration -- with the right notion of diversity -- is a practical path toward discovery of new behaviors beyond sharpening.

The most accurate monocular visual odometry systems require known camera intrinsics, refine their estimates with test-time optimization, and recover trajectories only up to an unknown factor. Systems built on large 3D models need no intrinsics, but they remain considerably less accurate and slower for odometry. Direct pose regression avoids all these requirements, yet it has not matched either approach's accuracy. We revisit this formulation with a transformer that predicts relative camera poses together with separate rotation and translation confidences over overlapping image windows, supervised by camera poses alone. A confidence-weighted module then aggregates the overlapping predictions into a single trajectory. The resulting method, CalfVO, needs no intrinsics, no bundle adjustment, and no loop closure, and it recovers scale from learned priors, accurately enough that it is evaluated without any alignment to the ground truth. Across five benchmarks, it is the most accurate calibration-free method on every metric we report, and it runs at 53 FPS, faster than every baseline.

The rise of large language models (LLMs) has led to dramatic improvements across a wide range of natural language tasks. Their performance on certain tasks can be further enhanced by incorporating test-time reasoning techniques. These inference-time advances have been adopted into the code domain, enabling complex software engineering (SWE) tasks such as code generation, test generation and issue resolution. However, the impact of different reasoning techniques on code-centric SWE tasks has not been systematically explored. In this work, we survey code reasoning techniques that underpin these capabilities, with a focus on test-time compute and inference-time reasoning paradigms. We examine a variety of code-specific reasoning methods and progressively build up to SWE agents, which combine planning, tool use, and multi-step interaction. We also compare the impact of different techniques on coding tasks, highlighting their relative importance and outlining open challenges and future research directions. Across commonly used models and benchmarks, we find that approaches exploiting code-specific signals (e.g., structure and execution feedback) are frequently associated with improved performance, motivating a dedicated study of code reasoning beyond natural-language reasoning.

As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.

Existing LLM-based agents have achieved strong performance on held-in tasks, but their generalizability to unseen tasks remains poor. Hence, some recent work focus on fine-tuning the policy model with more diverse tasks to improve the generalizability. In this work, we find that finetuning a reward model to guide the policy model is more robust than directly finetuning the policy model. Based on this finding, we propose AgentRM, a generalizable reward model, to guide the policy model for effective test-time search. We comprehensively investigate three approaches to construct the reward model, including explicit reward modeling, implicit reward modeling and LLM-as-a-judge. We then use AgentRM to guide the answer generation with Best-of-N sampling and step-level beam search. On four types of nine agent tasks, AgentRM enhances the base policy model by $8.8$ points on average, surpassing the top general agent by $4.0$. Moreover, it demonstrates weak-to-strong generalization, yielding greater improvement of $12.6$ on LLaMA-3-70B policy model. As for the specializability, AgentRM can also boost a finetuned policy model and outperform the top specialized agent by $11.4$ on three held-in tasks. Further analysis verifies its effectiveness in test-time scaling. Codes will be released to facilitate the research in this area.