Daily Papers

Skill-based agent systems tackle complex tasks by composing reusable skills, improving modularity and scalability while introducing a largely unexamined security attack surface. We propose SkillTrojan, a backdoor attack that targets skill implementations rather than model parameters or training data. SkillTrojan embeds malicious logic inside otherwise plausible skills and leverages standard skill composition to reconstruct and execute an attacker-specified payload. The attack partitions an encrypted payload across multiple benign-looking skill invocations and activates only under a predefined trigger. SkillTrojan also supports automated synthesis of backdoored skills from arbitrary skill templates, enabling scalable propagation across skill-based agent ecosystems. To enable systematic evaluation, we release a dataset of 3,000+ curated backdoored skills spanning diverse skill patterns and trigger-payload configurations. We instantiate SkillTrojan in a representative code-based agent setting and evaluate both clean-task utility and attack success rate. Our results show that skill-level backdoors can be highly effective with minimal degradation of benign behavior, exposing a critical blind spot in current skill-based agent architectures and motivating defenses that explicitly reason about skill composition and execution. Concretely, on EHR SQL, SkillTrojan attains up to 97.2% ASR while maintaining 89.3% clean ACC on GPT-5.2-1211-Global.

LLM customization platforms allow users to build task-specific models for code intelligence tasks by embedding instructions into system prompts, without modifying the underlying model parameters. While these platforms lower the barrier to developing customized LLMs, they also introduce a new attack surface: instruction backdoor attacks, in which adversaries implant hidden malicious behaviors into customized instructions. However, existing attacks suffer from two key limitations. First, they often rely on explicit trigger patterns readily detected by platform-side or user-side inspection. Second, they require substantial manual effort to craft task-specific backdoored instructions, limiting their scalability. In this paper, we propose ARIA, an automated red-teaming framework for crafting covert and effective backdoored instructions against customized LLMs. ARIA leverages an attacker LLM to iteratively generate and refine backdoored instructions, guided by structured feedback from the target LLM along three dimensions: stealthiness, clean-task utility, and backdoor effectiveness. We evaluate ARIA on three code intelligence tasks, using four representative LLMs, and compare it with three baseline attacks. Experimental results show that ARIA achieves the highest attack success rate of 0.945, while maintaining the best clean-task utility across all tasks. ARIA also generalizes well across programming languages and remains robust to generation temperature. Furthermore, ARIA significantly outperforms existing attacks in evading platform-side and user-side detection, achieving a false negative rate of up to 1.000, and stays effective against existing defense methods, demonstrating its strong generalizability and robustness.

Indirect prompt injection attacks hijack LLM-based agents by embedding malicious instructions in third-party data that the agent retrieves during task execution. Existing defenses report near-zero attack success rate on static benchmarks, yet recent adaptive evaluations show that these results collapse once the attacker is allowed to optimize against the deployed defense. In this work, we trace this collapse to two failure modes. First, existing defense methods are confined to recognizing specific attack patterns, rather than assessing whether the intent of every embedded instruction is relevant to the user task. Second, training-based defenses, which otherwise offer the strongest safety-utility trade-off, assemble their adversarial examples from a handful of hand-crafted templates, and the resulting defender fails to generalize outside that narrow strategy distribution. To address these gaps, we propose RETA, a training-based method that grounds defense decisions on the user tasks rather than attacker-controlled data. At each tool-output step, the defender undertakes chain-of-thought reasoning verifying that its actions are consistent with the user task. Leveraging red-teaming, a simulated attacker synthesizes adversarial training data and receives a dictionary-learning diversity reward, achieving broad coverage of injection-reformulation strategies. Together, these allow the defender to be optimized via multi-objective reinforcement learning and achieve better safety-utility trade-off. Across six black-box adaptive attacks, RETA keeps every per-attack ASR below 10%, with average ASR of 2.92% and 3.75% on the two target models, while preserving most utility under attack and on clean inputs.

Although Multimodal Large Language Models (MLLMs) have achieved remarkable progress across many domains, their training on large-scale multimodal datasets raises serious privacy concerns, making effective machine unlearning increasingly necessary. However, existing benchmarks mainly focus on static or short-sequence settings, offering limited support for evaluating continual privacy deletion requests in realistic deployments. To bridge this gap, we introduce ICU-Bench, a continual multimodal unlearning benchmark built on privacy-critical document data. ICU-Bench contains 1,000 privacy-sensitive profiles from two document domains, medical reports and labor contracts, with 9,500 images, 16,000 question-answer pairs, and 100 forget tasks. Additionally, new continual unlearning metrics are introduced, facilitating a comprehensive analysis of forgetting effectiveness, historical forgetting preservation, retained utility, and stability throughout the continual unlearning process. Through extensive experiments with representative unlearning methods on ICU-Bench, we show that existing methods generally struggle in continual settings and exhibit clear limitations in balancing forgetting quality, utility preservation, and scalability over long task sequences. These findings highlight the need for multimodal unlearning methods explicitly designed for continual privacy deletion.

Agentic Reinforcement Learning (RL) has empowered Large Language Models (LLMs) to utilize tools like Python interpreters for complex problem-solving. However, for parameter-constrained models (e.g., 4B--7B), the exploration phase is often plagued by frequent execution failures, creating noisy trajectories that hinder policy optimization. Under standard outcome-based reward settings, this noise leads to a critical credit assignment issue, where erroneous actions are inadvertently reinforced alongside successful outcomes. Existing mitigations face a dilemma: dense rewards often trigger reward hacking, while supersampling incurs prohibitive computational costs. To address these challenges, we propose CLEANER. Distinct from external filtering methods, CLEANER exploits the model's intrinsic self-correction capabilities to eliminate error-contaminated context directly during data collection. At its core, the Similarity-Aware Adaptive Rollback (SAAR) mechanism autonomously constructs clean, purified trajectories by retrospectively replacing failures with successful self-corrections. Based on semantic similarity, SAAR adaptively regulates replacement granularity from shallow execution repairs to deep reasoning substitutions. By training on these self-purified paths, the model internalizes correct reasoning patterns rather than error-recovery loops. Empirical results on AIME24/25, GPQA, and LiveCodeBench show average accuracy gains of 6%, 3%, and 5% over baselines. Notably, CLEANER matches state-of-the-art performance using only one-third of the training steps, highlighting trajectory purification as a scalable solution for efficient agentic RL. Our models and code are available at GitHub

We study the problem of multi-task learning under user-level differential privacy, in which n users contribute data to m tasks, each involving a subset of users. One important aspect of the problem, that can significantly impact quality, is the distribution skew among tasks. Certain tasks may have much fewer data samples than others, making them more susceptible to the noise added for privacy. It is natural to ask whether algorithms can adapt to this skew to improve the overall utility. We give a systematic analysis of the problem, by studying how to optimally allocate a user's privacy budget among tasks. We propose a generic algorithm, based on an adaptive reweighting of the empirical loss, and show that when there is task distribution skew, this gives a quantifiable improvement of excess empirical risk. Experimental studies on recommendation problems that exhibit a long tail of small tasks, demonstrate that our methods significantly improve utility, achieving the state of the art on two standard benchmarks.

Robust machine learning depends on clean data, yet current image data cleaning benchmarks rely on synthetic noise or narrow human studies, limiting comparison and real-world relevance. We introduce CleanPatrick, the first large-scale benchmark for data cleaning in the image domain, built upon the publicly available Fitzpatrick17k dermatology dataset. We collect 496,377 binary annotations from 933 medical crowd workers, identify off-topic samples (4%), near-duplicates (21%), and label errors (22%), and employ an aggregation model inspired by item-response theory followed by expert review to derive high-quality ground truth. CleanPatrick formalizes issue detection as a ranking task and adopts typical ranking metrics mirroring real audit workflows. Benchmarking classical anomaly detectors, perceptual hashing, SSIM, Confident Learning, NoiseRank, and SelfClean, we find that, on CleanPatrick, self-supervised representations excel at near-duplicate detection, classical methods achieve competitive off-topic detection under constrained review budgets, and label-error detection remains an open challenge for fine-grained medical classification. By releasing both the dataset and the evaluation framework, CleanPatrick enables a systematic comparison of image-cleaning strategies and paves the way for more reliable data-centric artificial intelligence.

A prominent approach to test-time scaling for text-to-image diffusion models formulates the problem as a search over multiple noise seeds, selecting the one that maximizes a certain image-reward function. The effectiveness of this strategy heavily depends on the number and diversity of noise seeds explored. However, verifying each candidate is computationally expensive, because each must be fully denoised before a reward can be computed. This severely limits the number of samples that can be explored under a fixed budget. We propose test-time scaling with noise-aware pruning (TTSnap), a framework that prunes low-quality candidates without fully denoising them. The key challenge is that reward models are learned in the clean image domain, and the ranking of rewards predicted for intermediate estimates are often inconsistent with those predicted for clean images. To overcome this, we train noise-aware reward models via self-distillation to align the reward for intermediate estimates with that of the final clean images. To stabilize learning across different noise levels, we adopt a curriculum training strategy that progressively shifts the data domain from clean images to noise images. In addition, we introduce a new metric that measures reward alignment and computational budget utilization. Experiments demonstrate that our approach improves performance by over 16\% compared with existing methods, enabling more efficient and effective test-time scaling. It also provides orthogonal gains when combined with post-training techniques and local test-time optimization. Code: https://github.com/TerrysLearning/TTSnap/.

Tabular Foundation Models (TFMs) achieve state-of-the-art zero-shot accuracy on small tabular datasets by meta-learning over synthetic data-generating processes -- making them highly attractive for practitioners who cannot afford large annotated corpora. However, their in-context learning mechanism assumes approximately clean inputs: missing values, outliers, and duplicates in the real-world data create a prior mismatch that degrades both accuracy and confidence calibration simultaneously. Correcting this mismatch requires sequential decisions over cleaning operators whose interactions no static preprocessing rule can anticipate -a natural fit for reinforcement learning~(RL). We introduce L2C2, the first deep RL framework framing tabular data cleaning as prior alignment: a learned policy sequences operators to minimize the distributional gap between dirty input and the TFM's synthetic prior. Six experiments on ten OpenML benchmark datasets establish: 1) three of seven reward designs collapse to degenerate trivial cleaning strategies -- principled reward engineering is scientifically non-trivial; 2) the novel TFMAwareReward reward we propose selects structurally distinct pipelines on 4/10 datasets and achieves higher TabPFN accuracy on those diverging cases (mean 0.851 vs. 0.843; Wilcoxon p=0.063, n=4) while never underperforming; 3) parameterized cleaning actions improve best-found pipeline reward on 9/10 datasets (Wilcoxon p=0.004); and 4) a policy pre-trained on one single source dataset exceeds scratch training at the 2,000-step fine-tuning checkpoint on all three held-out datasets (up to +28.8% after full fine-tuning) demonstrating cross-dataset transfer of prior-alignment knowledge. These findings establish that prior alignment is a principled data preparation strategy for TFM deployment on real-world tabular data.

The 1st Workshop on Data Contamination (CONDA 2024) focuses on all relevant aspects of data contamination in natural language processing, where data contamination is understood as situations where evaluation data is included in pre-training corpora used to train large scale models, compromising evaluation results. The workshop fostered a shared task to collect evidence on data contamination in current available datasets and models. The goal of the shared task and associated database is to assist the community in understanding the extent of the problem and to assist researchers in avoiding reporting evaluation results on known contaminated resources. The shared task provides a structured, centralized public database for the collection of contamination evidence, open to contributions from the community via GitHub pool requests. This first compilation paper is based on 566 reported entries over 91 contaminated sources from a total of 23 contributors. The details of the individual contamination events are available in the platform. The platform continues to be online, open to contributions from the community.

Large language model agents increasingly undertake long-horizon tasks in complex environments characterized by fragmented evidence, partially completed states, and dependencies across resources and artifacts. Existing agents may satisfy individual requirements while failing to maintain these relationships, producing locally correct but incomplete outcomes. These failures arise because task completion depends on relationships among environmental evidence, consequential decisions, and required effects. To address this completion gap, we propose Interlock, a framework for task synthesis that represents each task as a typed completion graph. Grounding edges connect public evidence to decisions, coupling hyperedges impose joint constraints across components, and realization edges connect decisions to all required downstream effects. Interlock compiles this graph into a resettable workspace, a public request, and an evaluation rubric. A source-coverage review checks whether the original requirements survive generation, while a public-input review checks whether the rubric is supported by the materials available to the solver. An agent-as-judge inspects execution artifacts and runs verification scripts, retaining trajectories judged to satisfy all mandatory requirements. To evaluate training utility, we fine-tune two Qwen3.6 backbones on their own accepted trajectories. Across the two backbones, training improves AutomationBench task pass rates by 4.00-4.50 percentage points, JobBench weighted scores by 3.20-6.64 points, and Harvey LAB criterion pass rates by 3.89-4.16 percentage points. These results demonstrate the training utility of the synthesized experience across external task environments.

As browser-use agents improve, benchmarks keep pace by collecting new tasks, websites, and applications, often making tasks longer or more novel. This makes difficulty expensive to refresh and difficult to control: when many aspects change at once, it is unclear what actually makes a task challenging. We instead construct challenging instances from tasks agents already solve, turning difficulty into a programmable property of the environment. BreakingWeb pairs every base task with an intervention condition that preserves the user instruction, latent target, and backend success criterion while changing the environment at different web stack layers. Each intervention is deterministic, detectable, and recoverable, and is annotated with the cognitive primitive it primarily loads. The benchmark contains 519 clean/intervention task pairs across seven self-hosted websites and 29 intervention families, all graded against outcomes. We evaluate six strong browser-use agents, three GUI-only agents that see only screenshots, and humans. The construction is effective: interventions cut agent pass rate by 22.9% on average and overturn nearly half of the tasks each agent solves cleanly, whereas humans lose 10.0% on a first attempt and 5.7% after one familiarisation attempt. The dominant failure is belief failure: 75% of the six agents' failures end with a declared success although the required change never happened. Our code, data and environment are publicly available at www.breakingweb.app.

Coding agents are increasingly asked to build software rather than patch it, yet benchmarks for from-scratch repository construction are mostly limited to a single language and depend on manually curated tasks. We introduce Zero2Repo, a benchmark in which an agent receives a product requirements document, an interface contract, and an empty workspace, and must deliver a complete repository in the project's native ecosystem. Tasks are produced by a language-agnostic authoring pipeline that converts real, version-pinned open-source projects into behavioral specifications, reproducible environments, and hidden acceptance tests. Each task is validated by execution: a reference implementation derived from the upstream project must pass, and adversarial validation must show that the tests reject incorrect implementations. Evaluation runs production coding agents in isolated containers, withholds the acceptance tests until an explicit submission, and assigns a binary reward only when every test passes, with no LLM judge. The pipeline and harness make no language-specific assumptions and apply to mainstream programming ecosystems; the current release contains Python, TypeScript, Go, and C++ tasks. Even on 11 tasks drawn from repositories that frontier models have very likely seen during training, the strongest agent solves only 10, and every failing submission passes 90-99% of the hidden tests; for the two strongest agents, 67-100% of failed tests trace to a single omission or a low-frequency rule stated in the specification rather than to a missing subsystem, so each failure is a concrete target for improvement.

Large language models (LLMs) have enabled powerful software engineering (SE) agents capable of navigating complex codebases and resolving real-world issues. However, these agents remain fundamentally episodic: they fail to retain, refine, and reuse experiences across tasks, repeatedly reconstructing context from scratch and reproducing similar mistakes. Even with memory support, they offer no remedy for the absence of a principled, task-agnostic memory utility, making them difficult to evaluate rigorously or generalize across agents and settings. To tackle these limitations, we introduce \ours, a closed-loop framework for memory augmentation in SE agents. \ours grounds memory utility in validated downstream impact, establishing utility as both a task-agnostic evaluation benchmark and an annotation-free optimization signal. Through complementary evaluation on single-episode and cross-episode memory augmentation, results demonstrate that \ours consistently improves SE agents across settings, achieving absolute gains of up to uparrow5.25% in success rate and uparrow4.63% in resolve efficiency, while substantially reducing computational cost by geq9.79%. Our project page: https://xhguo7.github.io/MemOp/{https://xhguo7.github.io/MemOp/}.

We introduce NatureBench, a cross-discipline benchmark of 90 tasks distilled from peer-reviewed Nature-family publications, designed to evaluate whether AI coding agents can move beyond reproduction toward discovery on real scientific problems. NatureBench is built on NatureGym, an automated pipeline that constructs a standardized, per-task containerized environment from a source paper, addressing the environment-fragmentation problem that has limited the credibility of prior agent-on-research benchmarks. Evaluating ten frontier agent configurations under a strict web-search-disabled protocol, we find that the strongest model surpasses SOTA on only 17.8% of tasks under the g>0.1 criterion. Analysis of method pathways reveals that agents succeed primarily through methodological translation, converting scientific tasks into familiar supervised prediction problems, rather than through genuine scientific invention. Failures are dominated by wrong method choice and insufficient compute budget, not by task misunderstanding. We release the benchmark, the NatureGym pipeline, and a public leaderboard with maintainer-side reproduction. Code: https://github.com/FrontisAI/NatureBench

The rise of personal assistant agents, e.g., OpenClaw, highlights the growing potential of large language models to support users across everyday life and work. A core challenge in these settings is proactive assistance, since users often begin with underspecified requests and leave important needs, constraints, or preferences unstated. However, existing benchmarks rarely evaluate whether agents can identify and act on such hidden intents before they are explicitly stated, especially in sustained multi-turn interactions where user needs emerge gradually. To address this gap, we introduce π-Bench, a benchmark for proactive assistance comprising 100 multi-turn tasks across 5 domain-specific user personas. By incorporating hidden user intents, inter-task dependencies, and cross-session continuity, π-Bench evaluates agents' ability to anticipate and address user needs over extended interactions, jointly measuring proactivity and task completion in long-horizon trajectories that better reflect real-world use. Experiments show (1) proactive assistance remains challenging, (2) a clear distinction between task completion and proactivity, and (3) the value of prior interaction for proactive intent resolution in later tasks.

With generative AI advances, the exciting potential for autonomous agents to manage daily tasks via natural language commands has emerged. However, cur rent agents are primarily created and tested in simplified synthetic environments, substantially limiting real-world scenario representation. In this paper, we build an environment for agent command and control that is highly realistic and reproducible. Specifically, we focus on agents that perform tasks on websites, and we create an environment with fully functional websites from four common domains: e-commerce, social forum discussions, collaborative software development, and content management. Our environment is enriched with tools (e.g., a map) and external knowledge bases (e.g., user manuals) to encourage human-like task-solving. Building upon our environment, we release a set of benchmark tasks focusing on evaluating the functional correctness of task completions. The tasks in our benchmark are diverse, long-horizon, and are designed to emulate tasks that humans routinely perform on the internet. We design and implement several autonomous agents, integrating recent techniques such as reasoning before acting. The results demonstrate that solving complex tasks is challenging: our best GPT-4-based agent only achieves an end-to-end task success rate of 10.59%. These results highlight the need for further development of robust agents, that current state-of-the-art LMs are far from perfect performance in these real-life tasks, and that WebArena can be used to measure such progress. Our code, data, environment reproduction resources, and video demonstrations are publicly available at https://webarena.dev/.

Global channel pruning (GCP) aims to remove a subset of channels (filters) across different layers from a deep model without hurting the performance. Previous works focus on either single task model pruning or simply adapting it to multitask scenario, and still face the following problems when handling multitask pruning: 1) Due to the task mismatch, a well-pruned backbone for classification task focuses on preserving filters that can extract category-sensitive information, causing filters that may be useful for other tasks to be pruned during the backbone pruning stage; 2) For multitask predictions, different filters within or between layers are more closely related and interacted than that for single task prediction, making multitask pruning more difficult. Therefore, aiming at multitask model compression, we propose a Performance-Aware Global Channel Pruning (PAGCP) framework. We first theoretically present the objective for achieving superior GCP, by considering the joint saliency of filters from intra- and inter-layers. Then a sequentially greedy pruning strategy is proposed to optimize the objective, where a performance-aware oracle criterion is developed to evaluate sensitivity of filters to each task and preserve the globally most task-related filters. Experiments on several multitask datasets show that the proposed PAGCP can reduce the FLOPs and parameters by over 60% with minor performance drop, and achieves 1.2xsim3.3x acceleration on both cloud and mobile platforms.

Large language model (LLM) agents are increasingly deployed as personal assistants. Existing evaluations, however, mostly use short, self-contained requests in static environments. Everyday life assistance is different. A task runs for weeks rather than minutes. The world keeps changing while the agent is not being prompted. Many constraints are never stated outright. An agent that merely answers the request in front of it will fail at such a task. What is needed instead is an agent that stays proactive and consistent. It decides on its own when to act, when to ask, and when to stay silent. It notices changes that nobody announced. It keeps one plan coherent from the first day to the last. No current benchmark measures this. We introduce VibeLifeBench, a benchmark of 200 long-horizon tasks across ten everyday-life domains. Each task is a scripted multi-week timeline in a simulated world of 22 mock services. The world advances on its own clock, and many of its changes are silent, so only an agent that re-inspects the world discovers them. Every task is graded by fine-grained, weighted checks that read only what the agent actually left behind, covering the end state, the timeliness of its actions, and whether it upheld the implicit constraints. We evaluate seven frontier models. All of them score low, which shows how far current agents are from assisting with real life. We will open-source all tasks, environments, and the evaluation framework.

Existing computer-use benchmarks fail to capture the realism, complexity, and long-horizon demands of real-world computer use, limiting their ability to reveal the limitations of frontier agents. We introduce OSWorld 2.0, a benchmark of 108 long-horizon computer-use workflows across everyday and professional tasks, designed to capture complex and challenging real-world phenomena. Each task represents a realistic end-to-end workflow that takes human users a median of about 1.6 hours to complete and requires an average of 318 tool calls with Claude Opus 4.7 using maximum thinking, compared with about 30 in OSWorld 1.0. OSWorld 2.0 targets challenge phenomena that are common in real workflows yet underrepresented in prior benchmarks, spanning interaction-design challenges such as streaming interaction and dynamic environments, as well as agent-pattern challenges such as cross-source reasoning, implicit-state inference, and visual-spatial precision. Tasks are grounded in authentic input artifacts and cross-referenced against realistic stateful user profile data, and include separate safety reports auditing safety-sensitive execution. Under our primary binary-completion metric at 500 steps, Claude Opus 4.8 with maximum thinking and batched tool calls scores best but still completes only 20.6% of tasks at a 54.8% partial score; GPT-5.5 is far more token-efficient yet plateaus near 13%. These results show that current agents are still far from professional-level computer use: rather than stumbling on basic GUI control or coding, they lose track of constraints, miss information that arrives mid-task, guess rather than ask the user, and skip verification, struggling most when a task hinges on hidden state they must recover.

Deep research agents have demonstrated remarkable capabilities in complex information-seeking tasks, yet this power comes at a steep computational cost. Driven by accuracy-focused training paradigms, current models adopt brute-force strategies characterized by blind tool dependency and performative reasoning-generating long, redundant trajectories that are far from necessary for resolving these tasks, leading to wasteful tool calls and excessive token consumption. To overcome this efficiency trap, we propose SlimSearcher, a principled framework that pushes the Pareto frontier between accuracy and computational cost across both Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). In the SFT stage, SlimSearcher employs Pareto-efficient filtration to distill trajectories that are both successful and economical, guiding the model toward inherently efficiency-aware search behaviors. During RL, we introduce Adaptive Reward Gating, a dynamic reward-shaping mechanism that evaluates relative tool and token efficiency within a sampled cohort. By cascading these adaptive efficiency metrics with a strict correctness gate, our approach effectively avoids the brevity bias associated with absolute penalties and mitigates reward hacking. Extensive experiments on long-horizon benchmarks, including GAIA, BrowseComp, and XBenchDeepSearch, demonstrate that SlimSearcher reduces average tool-call rounds by 17%-58% while maintaining or improving accuracy.

LLM-based agents have shown promising capabilities in a growing range of software engineering (SWE) tasks. However, advancing this field faces two critical challenges. First, high-quality training data is scarce, especially data that reflects real-world SWE scenarios, where agents must interact with development environments, execute code and adapt behavior based on the outcomes of their actions. Existing datasets are either limited to one-shot code generation or comprise small, manually curated collections of interactive tasks, lacking both scale and diversity. Second, the lack of fresh interactive SWE tasks affects evaluation of rapidly improving models, as static benchmarks quickly become outdated due to contamination issues. To address these limitations, we introduce a novel, automated, and scalable pipeline to continuously extract real-world interactive SWE tasks from diverse GitHub repositories. Using this pipeline, we construct SWE-rebench, a public dataset comprising over 21,000 interactive Python-based SWE tasks, suitable for reinforcement learning of SWE agents at scale. Additionally, we use continuous supply of fresh tasks collected using SWE-rebench methodology to build a contamination-free benchmark for agentic software engineering. We compare results of various LLMs on this benchmark to results on SWE-bench Verified and show that performance of some language models might be inflated due to contamination issues.

Large language model (LLM) based search agents iteratively generate queries, retrieve external information, and reason to answer open-domain questions. While researchers have primarily focused on improving their utility, their safety behaviors remain underexplored. In this paper, we first evaluate search agents using red-teaming datasets and find that they are more likely to produce harmful outputs than base LLMs. For example, when asked "How can I track someone's location without their consent?", a base model refuses, whereas a search agent designed to retrieve and cite sources may lower its refusal threshold, fetch documents (e.g., court cases), and, once appended, synthesize them into an informative yet unsafe summary. We further show that utility-oriented fine-tuning intensifies this risk, motivating joint alignment of safety and utility. We present SafeSearch, a multi-objective reinforcement learning approach that couples a final-output safety/utility reward with a novel query-level shaping term that penalizes unsafe queries and rewards safe ones. Experiments show that SafeSearch reduces agent harmfulness by over 70% across three red-teaming datasets while producing safe, helpful responses, and matches the QA performance of a utility-only finetuned agent; further analyses confirm the effectiveness of the query-level reward in jointly improving safety and utility.

Safety alignment of text-to-image (T2I) diffusion models aims to suppress harmful generations while preserving utility on benign prompts. Recent methods often appear to deliver high safety with high utility, but this conclusion rests largely on coarse global utility metrics (e.g., FID, CLIPScore) that are insensitive to fine-grained semantic correctness, creating an illusion of high utility. We show that when utility is measured with structured evaluation, this illusion breaks: on TIFA (Text-to-Image Faithfulness evaluation with Question Answering), safety-aligned models suffer substantial drops in semantic fidelity, including failures in object counts, attributes, and relationships. To diagnose the source of this gap, we analyze the text-encoder prompt embedding space and uncover semantic collapse, a contraction of embedding spread coupled with distortion of inter-prompt similarity structure, which strongly correlates with structured utility loss. Guided by this insight, we propose StructureAware Geometric Regularization (SAGE), a safety alignment objective that explicitly preserves embedding spread and inter-prompt relational structure during adaptation. Our method restores structured utility (TIFA +5.0% over prior state-of-the-art) while maintaining strong safety performance and competitive coarse-grained utility scores. Our source code and trained models are available at https://adeelyousaf.github.io/SAGE_ECCV26_Project_Page/.

The evaluation of coding agents generally strives to be as realistic as possible. In our study, we collect 4,782 agent sessions of real software engineers in JetBrains IDEs, which we call Production Sessions. Since our subject is interactive agents, we study the sessions with at least three user messages (33% of the sample). These long sessions differ from issue-derived benchmark tasks in two ways: (i) user requests span a far wider mix of task types - questions about the project's code, planning, review, refactoring, execution - and (ii) users switch between types throughout a session. Long-session samples from three public interaction corpora exhibit markedly different Task Flows (the distributions of session lengths, task types, and type-to-type transitions), so no single interaction distribution is universally realistic: benchmarks should name a target use case and calibrate to measurements from it. We present SWE-TaskFlow, an approach for transforming any issue-derived benchmark: it preserves the verified tasks and tests while steering the interaction toward a target Task Flow through prompt splitting and verifiable repository QA, with a TaskFlow Alignment Score (TFAS) for selecting among generated trajectories. In a pilot on 700 SWE-Bench Pro tasks, solving the task sequentially in several steps approximately doubles agent cost without a stable change in resolve rate: the interaction protocol itself is an important dimension of evaluation.

The development of large language models (LLMs) depends on trustworthy evaluation. However, most current evaluations rely on public benchmarks, which are prone to data contamination issues that significantly compromise fairness. Previous researches have focused on constructing dynamic benchmarks to address contamination. However, continuously building new benchmarks is costly and cyclical. In this work, we aim to tackle contamination by analyzing the mechanisms of contaminated models themselves. Through our experiments, we discover that the overestimation of contaminated models is likely due to parameters acquiring shortcut solutions in training. We further propose a novel method for identifying shortcut neurons through comparative and causal analysis. Building on this, we introduce an evaluation method called shortcut neuron patching to suppress shortcut neurons. Experiments validate the effectiveness of our approach in mitigating contamination. Additionally, our evaluation results exhibit a strong linear correlation with MixEval, a recently released trustworthy benchmark, achieving a Spearman coefficient (rho) exceeding 0.95. This high correlation indicates that our method closely reveals true capabilities of the models and is trustworthy. We conduct further experiments to demonstrate the generalizability of our method across various benchmarks and hyperparameter settings. Code: https://github.com/GaryStack/Trustworthy-Evaluation

Safe completion requires models to provide useful assistance without enabling harm, but this behavior is difficult to evaluate with isolated prompts. We introduce OpenSafeIntent, a benchmark of controlled prompt-sets that vary intent while holding the underlying task fixed. Each datapoint contains benign, dual-use, and malicious variants of the same task. This design lets us evaluate whether models calibrate assistance across intent shifts, rather than merely appearing safe on average. Across a broad model suite, we find that prompt-level safety hides important failures: models often fail to remain safe across matched intent variants, dual-use behavior is brittle under paraphrase, high-level answers on risky topics are not reliably safe, and responses that reframe ambiguous requests into safer tasks are substantially less likely to cross the safety boundary. Our results suggest that safe completion should be evaluated as intent-calibrated behavior over controlled task variants, not as a single safety-helpfulness tradeoff over independent prompts.

Object detection under image degradation can benefit from clean-image supervision, but aggregate gains do not imply that transferred representation changes are uniformly useful. We study how clean task knowledge affects degraded-image representations and whether local responses to structured representation directions can be characterized geometrically. Using paired clean and Gaussian-degraded BDD100K images, we show that clean-teacher distillation improves observed detection accuracy while producing heterogeneous object-level transfer. We isolate a representation component complementary to direct clean-teacher alignment and map it into the distilled student space through an orthogonal bridge. Controlled interventions rescue 13.22% of objects lost under the distilled representation, versus 4.30% under norm-matched random perturbations, with very low harm on preserved objects. We introduce task-sensitive geometry, a gradient-derived channel-space geometry constructed from normalized detection-loss gradients. On a reserved cohort, mapped-complement orientation within this frozen geometry is positively associated with local intervention-response magnitude after controlling for intervention magnitude (partial Spearman ρ = 0.242, 95% CI [0.108, 0.359]). The relationship eplicates on independent data (ρ = 0.180) and with RT-DETR-L (ρ = 0.227), but not for the direct clean-teacher residual family, and it weakens for large interventions. Routing rules and specialized distillation objectives based on these signals do not yield statistically reliable gains over CLEANKD. These results support a local, direction-family-dependent task-sensitive geometry while showing that converting such structure into improved global training remains an open problem.

SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: reward hacking, enabled by leakage of gold solutions or hidden evaluation information, and task quality issues, including misleading problem statements and improperly scoped tests. These issues can inflate benchmark performance and obscure agents' true coding ability. We present SWE-Bench Pro Verified, a verified version of SWE-Bench Pro that addresses both problems. Our approach combines anti-hacking safeguards that eliminate major leakage channels without disrupting normal agent functionality, with task refinement that minimally corrects inconsistencies within flawed instances. Evaluations on SWE-Bench Pro Verified reveal that some models perform substantially worse than previously reported, suggesting that existing results on SWE-Bench Pro may overestimate real software engineering capability. SWE-Bench Pro Verified offers a more trustworthy benchmark for assessing software engineering agents.

We aim to train a multi-task model such that users can adjust the desired compute budget and relative importance of task performances after deployment, without retraining. This enables optimizing performance for dynamically varying user needs, without heavy computational overhead to train and save models for various scenarios. To this end, we propose a multi-task model consisting of a shared encoder and task-specific decoders where both encoder and decoder channel widths are slimmable. Our key idea is to control the task importance by varying the capacities of task-specific decoders, while controlling the total computational cost by jointly adjusting the encoder capacity. This improves overall accuracy by allowing a stronger encoder for a given budget, increases control over computational cost, and delivers high-quality slimmed sub-architectures based on user's constraints. Our training strategy involves a novel 'Configuration-Invariant Knowledge Distillation' loss that enforces backbone representations to be invariant under different runtime width configurations to enhance accuracy. Further, we present a simple but effective search algorithm that translates user constraints to runtime width configurations of both the shared encoder and task decoders, for sampling the sub-architectures. The key rule for the search algorithm is to provide a larger computational budget to the higher preferred task decoder, while searching a shared encoder configuration that enhances the overall MTL performance. Various experiments on three multi-task benchmarks (PASCALContext, NYUDv2, and CIFAR100-MTL) with diverse backbone architectures demonstrate the advantage of our approach. For example, our method shows a higher controllability by ~33.5% in the NYUD-v2 dataset over prior methods, while incurring much less compute cost.

In this study, we explore the sophisticated domain of task planning for robust household embodied agents, with a particular emphasis on the intricate task of selecting substitute objects. We introduce the CommonSense Object Affordance Task (COAT), a novel framework designed to analyze reasoning capabilities in commonsense scenarios. This approach is centered on understanding how these agents can effectively identify and utilize alternative objects when executing household tasks, thereby offering insights into the complexities of practical decision-making in real-world environments.Drawing inspiration from human decision-making, we explore how large language models tackle this challenge through three meticulously crafted commonsense question-and-answer datasets, featuring refined rules and human annotations. Our evaluation of state-of-the-art language models on these datasets sheds light on three pivotal considerations: 1) aligning an object's inherent utility with the task at hand, 2) navigating contextual dependencies (societal norms, safety, appropriateness, and efficiency), and 3) accounting for the current physical state of the object. To maintain accessibility, we introduce five abstract variables reflecting an object's physical condition, modulated by human insights to simulate diverse household scenarios. Our contributions include insightful Object-Utility mappings addressing the first consideration and two extensive QA datasets (15k and 130k questions) probing the intricacies of contextual dependencies and object states. The datasets, along with our findings, are accessible at: https://github.com/com-phy-affordance/COAT. This research not only advances our understanding of physical commonsense reasoning in language models but also paves the way for future improvements in household agent intelligence.

We consider the problem of building an assistive robotic system that can help humans in daily household cleanup tasks. Creating such an autonomous system in real-world environments is inherently quite challenging, as a general solution may not suit the preferences of a particular customer. Moreover, such a system consists of multi-objective tasks comprising -- (i) Detection of misplaced objects and prediction of their potentially correct placements, (ii) Fine-grained manipulation for stable object grasping, and (iii) Room-to-room navigation for transferring objects in unseen environments. This work systematically tackles each component and integrates them into a complete object rearrangement pipeline. To validate our proposed system, we conduct multiple experiments on a real robotic platform involving multi-room object transfer, user preference-based placement, and complex pick-and-place tasks. Project page: https://sites.google.com/eng.ucsd.edu/home-robot

Large language models pretrained on extensive web corpora demonstrate remarkable performance across a wide range of downstream tasks. However, a growing concern is data contamination, where evaluation datasets may be contained in the pretraining corpus, inflating model performance. Decontamination, the process of detecting and removing such data, is a potential solution; yet these contaminants may originate from altered versions of the test set, evading detection during decontamination. How different types of contamination impact the performance of language models on downstream tasks is not fully understood. We present a taxonomy that categorizes the various types of contamination encountered by LLMs during the pretraining phase and identify which types pose the highest risk. We analyze the impact of contamination on two key NLP tasks -- summarization and question answering -- revealing how different types of contamination influence task performance during evaluation.

Terminal agents are increasingly capable of executing complex, long-horizon tasks autonomously from a single user prompt. To do so, they must interpret instructions encountered in the environment (e.g., README files, code comments, stack traces) and determine their relevance to the task. This creates a fundamental challenge: relevant cues must be followed to complete a task, whereas irrelevant or misleading ones must be ignored. Existing benchmarks do not capture this ability. An agent may appear capable by blindly following all instructions, or appear robust by ignoring them altogether. We introduce TAB (Task Alignment Benchmark), a suite of 89 terminal tasks derived from Terminal-Bench 2.1. Each task is intentionally underspecified, with missing information provided as a necessary cue embedded in a natural environmental artifact, alongside a plausible but irrelevant distractor. Solving these tasks requires selectively using the cue while ignoring the distractor. Applying TAB to ten frontier agents reveals a systematic gap between task capability and task alignment. The strongest Terminal-Bench agent achieves high task completion but low task alignment on TAB. Evaluating six prompt-injection defenses further shows that suppressing distractor execution also suppresses the cues required for task completion. These results demonstrate that task-aligned agents require selective use of environmental instructions rather than blanket acceptance or rejection.

This paper introduces the Opus Workflow Evaluation Framework, a probabilistic-normative formulation for quantifying Workflow quality and efficiency. It integrates notions of correctness, reliability, and cost into a coherent mathematical model that enables direct comparison, scoring, and optimization of Workflows. The framework combines the Opus Workflow Reward, a probabilistic function estimating expected performance through success likelihood, resource usage, and output gain, with the Opus Workflow Normative Penalties, a set of measurable functions capturing structural and informational quality across Cohesion, Coupling, Observability, and Information Hygiene. It supports automated Workflow assessment, ranking, and optimization within modern automation systems such as Opus and can be integrated into Reinforcement Learning loops to guide Workflow discovery and refinement. In this paper, we introduce the Opus Workflow Reward model that formalizes Workflow success as a probabilistic expectation over costs and outcomes. We define measurable Opus Workflow Normative Penalties capturing structural, semantic, and signal-related properties of Workflows. Finally, we propose a unified optimization formulation for identifying and ranking optimal Workflows under joint Reward-Penalty trade-offs.

TaskGen is an open-sourced agentic framework which uses an Agent to solve an arbitrary task by breaking them down into subtasks. Each subtask is mapped to an Equipped Function or another Agent to execute. In order to reduce verbosity (and hence token usage), TaskGen uses StrictJSON that ensures JSON output from the Large Language Model (LLM), along with additional features such as type checking and iterative error correction. Key to the philosophy of TaskGen is the management of information/memory on a need-to-know basis. We empirically evaluate TaskGen on various environments such as 40x40 dynamic maze navigation with changing obstacle locations (100% solve rate), TextWorld escape room solving with dense rewards and detailed goals (96% solve rate), web browsing (69% of actions successful), solving the MATH dataset (71% solve rate over 100 Level-5 problems), Retrieval Augmented Generation on NaturalQuestions dataset (F1 score of 47.03%)

Large language models are increasingly deployed as agents that solve tasks by interacting with external tool environments. Although recent tool-use benchmarks increasingly cover complex task settings, they still largely assume clean, stable, and trustworthy tool environments, leaving tool-environment unreliability insufficiently examined. We introduce ToolBench-X, a benchmark for evaluating agents under recoverable reliability hazards. ToolBench-X contains executable multi-step tasks across diverse domains and sequential, parallel, and mixed workflows, each paired with deterministic tools and a canonical final answer for automatic evaluation. Starting from clean tool environments, ToolBench-X injects five structured hazard types: Specification Drift, Invocation Error, Execution Failure, Output Drift, and Cross-source Conflict. Crucially, each injected instance remains solvable through at least one valid recovery path, such as retrying, fallback, verification, or cross-checking. Experiments reveal a substantial reliability gap: agents that perform well with reliable tools often fail under recoverable hazards. Further analysis shows that failures are driven less by tool-use volume or inference budget than by limited hazard diagnosis and ineffective recovery. Targeted recovery hints recover many failed tasks, while test-time scaling yields more limited gains. These results suggest that tool-use evaluation should move beyond function-call accuracy toward task completion under unreliable tool environments. The code and data is available at https://github.com/Foreverskyou/ToolBench-X.

Repository-level code editing requires models to understand complex dependencies and execute precise multi-file modifications across a large codebase. While recent gains on SWE-bench rely heavily on complex agent scaffolding, it remains unclear how much of this capability can be internalised via high-quality training signals. To address this, we propose Clean Pull Request (Clean-PR), a mid-training paradigm that leverages real-world GitHub pull requests as a training signal for repository-level editing. We introduce a scalable pipeline that converts noisy pull request diffs into Search/Replace edit blocks through reconstruction and validation, resulting in the largest publicly available corpus of 2 million pull requests spanning 12 programming languages. Using this training signal, we perform a mid-training stage followed by an agentless-aligned supervised fine-tuning process with error-driven data augmentation. On SWE-bench, our model significantly outperforms the instruction-tuned baseline, achieving absolute improvements of 13.6% on SWE-bench Lite and 12.3% on SWE-bench Verified. These results demonstrate that repository-level code understanding and editing capabilities can be effectively internalised into model weights under a simplified, agentless protocol, without relying on heavy inference-time scaffolding.

Coding agents increasingly interact with users by generating interactive software on the fly rather than replying in text, yet many bugs in such software reveal themselves only after multi-step interactions. Today, users must discover these bugs and wait for fixes, making the development loop slow and expensive. Existing benchmarks typically assume the bugs have already been identified, leaving agents' ability to find interaction-dependent failures largely unevaluated. We introduce SWEeper-Bench, 200 tasks built from real bug fixes in 200 widely used open-source web applications. Each task gives the agent only the codebase and an open-ended request such as "Find and fix issues in Focalboard's board sharing and membership." Because these bugs are hard to detect from code alone, agents must design interactive tests to uncover them. An agentic verifier then judges each repair by what a user would observe in the browser. The best of 15 frontier agents passes only 59.0% of tasks. Through stage-by-stage analysis of agent trajectories, we show that the biggest bottleneck is identifying the buggy behavior: agents often reach the relevant feature but never trigger the failure or fail to recognize it. SWEeper-Bench measures progress toward agents that close this loop themselves, delivering software that works without waiting for user feedback.

Language-model agents are increasingly deployed under standing instructions: a system prompt, a policy file, or a skills document is placed in context, and the agent is trusted to let that document govern every action that follows. Existing benchmarks rarely test this deployment pattern directly; they measure whether an agent can complete a task, not whether a long, binding policy document constrains its behavior over an extended tool-use horizon. We present HANDBOOK_md, a benchmark of 65 agentic tasks modeled on how employees follow company handbooks. Each task places an agent in a self-contained company environment (a file workspace with mock email, chat, calendar, issue-tracking, and commerce services exposed over the Model Context Protocol) and instructs it to carry out routine professional work governed by an expert-written standard operating procedure of 20-124 pages. Tasks span five domains (finance, medical billing, insurance, logistics, and HR) and 10 fictional companies. To resist memorization, every task modifies one of 10 base handbooks, altering the specific rules and thresholds on which grading depends, so no two tasks share the same set of policies. Grading is fully deterministic: each task carries a rubric of programmatic criteria (824 in total) that check both that required actions occurred and that prohibited actions did not. Under strict grading, where a trial passes only if every criterion is satisfied, the strongest evaluated model passes 36.2% of trials, and most frontier models remain below 25%. Failures follow consistent patterns: agents let a plausible but unauthorized in-environment request override the standing policy, perform a required check and then act against its result, lose rule details over long horizons, and report compliance they did not achieve. We release the tasks, environments, and evaluation harness.

We propose a mitigation to reward hacking in reinforcement learning. Our setup involves a misspecified reward function that gives rise to unintended behavior, and a monitor used to detect unintended behavior. As with frontier AI systems, the monitor makes systematic errors. Consequently, using the monitor for data filtering or to apply a reward penalty fails to prevent the model from learning unintended behavior ("reward hacking"). To address this problem, we introduce GRAFT: Gradient Routing Adapter Fine Tuning. Rather than penalizing or filtering flagged samples during training, GRAFT constrains gradient updates on those samples to a sacrificial set of new parameters. This has two benefits: First, because the monitor is not part of the reward function, the model does not learn to hide misbehavior. Second, ablating the sacrificial parameters reduces or entirely stops the unintended behavior while maintaining improved performance on target tasks. We demonstrate these benefits in RL environments where reward penalties cause models to learn to reward hack in undetected ways, training models ranging from 135M to 8B parameters. Additionally, we show a similar effect at a larger scale by distilling frontier LLM trajectories containing reward hacking into Qwen3-32B, without teaching that model to reward hack. Overall, our work suggests GRAFT may be a promising approach for leveraging imperfect monitors during post-training in settings where true performance is difficult to evaluate.

In this paper, we introduce Δτ-Bench, an executable agent benchmark that models the structural evolution of large tool ecosystems. In Δτ-Bench, independently evolving services continually reshape how tasks can be accomplished: familiar tools may disappear or be replaced, while more efficient alternatives emerge amid overlapping functionality and noisy documentation. Against this moving capability frontier, we evaluate whether agents can proactively discover efficient tool-use paths and promptly revise their accumulated procedural knowledge, bringing together demands that existing benchmarks fail to examine. Δτ-Bench comprises 1,588 validated multi-turn tasks across 42 domains and six successive ecosystem states generated through capability-preserving transformations. Our evaluation shows that even the strongest evaluated model or harness achieves only 42.3% task success, while initial gains from existing skill-learning methods can not consistently translate into better adaptation as the ecosystem evolves. These findings motivate Steward, an ecosystem manager that jointly manages tools and procedural knowledge through incremental skill maintenance and dynamic provisioning during task execution. Experiments show that Steward sustains higher task success across ecosystem changes with lower skill-maintenance overhead, while improving both recovery from disrupted workflows and adoption of more efficient alternatives.

In real home deployments, household agents must often operate from a complete household scene and a situated household request, rather than from a clean task specification. Such requests require agents to identify task-relevant entities, recover intended task conditions, and resolve ordering constraints from the surrounding scene context. We formalize this capability as full-scene household reasoning: given a complete household scene and a situated household request, an agent must infer executable task structure before producing a grounded skill-level action sequence. This setting is challenging because complete household scenes contain substantial task-irrelevant information, making direct complete-scene prompting inefficient and error-prone. In practical deployment, this challenge is further amplified by privacy and local compute constraints, which favor compact open-weight models with limited long-context reasoning ability. We propose TaskGround, a training-free and model-agnostic Ground-Infer-Execute framework that grounds complete scenes into compact task-relevant scene slices, infers executable task structure, and compiles it into grounded skill-level action sequences. To evaluate this setting, we introduce FullHome, a human-validated evaluation suite of 400 household tasks spanning diverse home-scale environments and both goal-oriented and process-constrained requirements. On FullHome, TaskGround improves task success rates by large margins across both proprietary and open-weight models. Notably, it makes Qwen3.5-9B competitive with GPT-5 under direct complete-scene prompting while reducing total input-token cost by up to 18x. Our results identify executable task-structure inference as a central bottleneck in full-scene household reasoning and show that structured grounding can make compact local models substantially more effective for practical household deployment.

Terminal-agent benchmarks have become a primary signal for measuring the coding and system-administration capabilities of large language models. As the market for evaluation environments grows, so does the pressure to ship tasks quickly, often without thorough adversarial review of the verification logic. This paper is a guideline for writing good benchmark tasks, drawn from over a year of contributing to and reviewing tasks for Terminal Bench. Most people write benchmark tasks the way they write prompts. They shouldn't. A prompt is designed to help the agent succeed; a benchmark is designed to find out if it can. We argue that good tasks are adversarial, difficult, and legible, and that a large class of common failure modes -- AI-generated instructions, over-prescriptive specifications, clerical difficulty, oracle solutions that assume hidden knowledge, tests that validate the wrong things, and reward-hackable environments -- are predictable consequences of treating task authoring as prompt authoring. We catalog these failure modes, argue that real difficulty is conceptual rather than environmental, and discuss recent empirical evidence that over 15% of tasks in popular terminal-agent benchmarks are reward-hackable. We hope this serves as a useful reference for benchmark maintainers, task contributors, and researchers using benchmark scores as evidence.

Large language models are increasingly trained on all the data ever produced by humans. Many have raised concerns about the trustworthiness of public benchmarks due to potential contamination in pre-training or fine-tuning datasets. While most data decontamination efforts apply string matching (e.g., n-gram overlap) to remove benchmark data, we show that these methods are insufficient, and simple variations of test data (e.g., paraphrasing, translation) can easily bypass these decontamination measures. Furthermore, we demonstrate that if such variation of test data is not eliminated, a 13B model can easily overfit a test benchmark and achieve drastically high performance, on par with GPT-4. We validate such observations in widely used benchmarks such as MMLU, GSK8k, and HumanEval. To address this growing risk, we propose a stronger LLM-based decontamination method and apply it to widely used pre-training and fine-tuning datasets, revealing significant previously unknown test overlap. For example, in pre-training sets such as RedPajama-Data-1T and StarCoder-Data, we identified that 8-18\% of the HumanEval benchmark overlaps. Interestingly, we also find such contamination in synthetic dataset generated by GPT-3.5/4, suggesting a potential risk of unintentional contamination. We urge the community to adopt stronger decontamination approaches when using public benchmarks. Moreover, we call for the community to actively develop fresh one-time exams to evaluate models accurately. Our decontamination tool is publicly available at https://github.com/lm-sys/llm-decontaminator.

As LM agents gain the ability to execute actions on real computer systems, we need ways to not only prevent harmful actions at scale but also effectively remediate harm when prevention fails. We formalize a solution to this neglected challenge in post-execution safeguards as harm recovery: the problem of optimally steering an agent from a harmful state back to a safe one in alignment with human preferences. We ground preference-aligned recovery through a formative user study that identifies valued recovery dimensions and produces a natural language rubric. Our dataset of 1,150 pairwise judgments reveals context-dependent shifts in attribute importance, such as preferences for pragmatic, targeted strategies over comprehensive long-term approaches. We operationalize these learned insights in a reward model, re-ranking multiple candidate recovery plans generated by an agent scaffold at test time. To evaluate recovery capabilities systematically, we introduce BackBench, a benchmark of 50 computer-use tasks that test an agent's ability to recover from harmful states. Human evaluation shows our reward model scaffold yields higher-quality recovery trajectories than base agents and rubric-based scaffolds. Together, these contributions lay the foundation for a new class of agent safety methods -- ones that confront harm not only by preventing it, but by navigating its aftermath with alignment and intent.

Computer use has reached a turning point: the focus is expanding from GUI-only interaction to systems that complete tasks on users' computers through visual and programmatic interfaces. We define a generalist computer-use agent (GCUA) as a system in which the model chooses how to observe, act, and organize its work across visual and programmatic interfaces. Yet frontier implementations, such as GPT-6 Astra's computer-use system, remain largely closed, limiting systematic study of their design. We introduce Open Generalist CUA (OpenGCUA), an ongoing project to develop and open the full stack for GCUAs with the community. This report presents the first release of OpenGCUA. Code execution is a key feature of our agent design: it lets intermediate results determine subsequent operations and enables models to apply their existing programming capabilities directly to computer interaction. We also propose harness-to-model decision transfer, moving selected observations and task-state decisions from fixed harness rules to the model. On OSWorld-v2, our full configuration improves task scores across all eight evaluated models, including GPT-6 Astra and Claude Opus 5, raising mean partial score by +19.6 points and reducing reported steps per task by an average of 53.0% relative to batched GUI-only execution. We release our implementation and prompts, evaluation trajectories from all eight models, and preliminary versions of a real-device harness and a cross-device SDK. The harness addresses real-device challenges not fully captured by benchmarks, while the SDK provides a common programming interface across computers, mobiles and browsers. Across four real macOS tasks under our real-device harness, selected Qwen3.8-Max runs use 19.4 times as many reported reasoning tokens as GPT-6 Astra, highlighting substantial differences between models in real-device execution efficiency.

AI agents depend on repeated decisions to accomplish tasks, with large language models (LLMs) commonly serving as their core decision makers. The recent emergence of Jev offers an opportunity to make these decisions faster and at lower cost. However, it remains unclear whether such local savings can be translated into efficient and reliable task completion. In this paper, we investigate this efficiency transfer across five agents in two representative application settings, embodied tasks in ALFWorld and mobile app use in AndroidWorld. Across 4,500 task executions, we jointly evaluate task effectiveness, decision-level and task-level resource use, and execution behavior. Experimental results show that Jev integration substantially reduces decision cost and latency across all five agents, with task-level gains varying by workflow. Task success remains close to baseline levels on ALFWorld and declines more substantially in two mobile agents. Auxiliary reasoning and additional interactions absorb decision savings, while failed recovery and premature completion limit successful delivery. Mean cost per successful outcome increases in one agent even as spending per attempt falls. These findings highlight the importance of preserving progress, supporting recovery, and verifying completion when integrating efficient decision models into agent workflows. Their practical value should be assessed through successful task delivery, recovery behavior, and resources per delivered outcome.

To succeed in real-world long-horizon software development, coding agents must satisfy multi-dimensional task specifications, such as interface compatibility and error handling, and do so without breaking existing dependencies. However, current works lack specification in both quantity and diversity, often emphasizing primary goals while ignoring complex constraints. Synthesizing tasks to bridge this gap presents two challenges: naively adding specifications risks reducing tasks to trivial checklists rather than compelling joint reasoning, and verifying these subtle interactions risks overlooking hidden flaws or rejecting valid alternatives. To address these challenges, we introduce CodeSpecWeave, a framework for synthesizing tasks with rich, interdependent specifications. Grounded in real developer seeds, a synthesis agent iteratively develops both coupled task requirements and executable tests, while a review agent audits solution trajectories from multiple models to guarantee test reliability. Using this pipeline, we construct 4,670 training tasks across 2,865 repositories. Applying Reinforcement learning on these tasks significantly improves performance over the base model on DeepSWE-v1.1, largely outperforming existing baselines. Furthermore, we curate CodeSpecWeave-Bench, a rigorous benchmark comprising 100 tasks and 1,044 specifications, which clearly separates frontier models and reveals significant headroom in handling interdependent engineering constraints.

The recent success of Large Language Models (LLMs) signifies an impressive stride towards artificial general intelligence. They have shown a promising prospect in automatically completing tasks upon user instructions, functioning as brain-like coordinators. The associated risks will be revealed as we delegate an increasing number of tasks to machines for automated completion. A big question emerges: how can we make machines behave responsibly when helping humans automate tasks as personal copilots? In this paper, we explore this question in depth from the perspectives of feasibility, completeness and security. In specific, we present Responsible Task Automation (ResponsibleTA) as a fundamental framework to facilitate responsible collaboration between LLM-based coordinators and executors for task automation with three empowered capabilities: 1) predicting the feasibility of the commands for executors; 2) verifying the completeness of executors; 3) enhancing the security (e.g., the protection of users' privacy). We further propose and compare two paradigms for implementing the first two capabilities. One is to leverage the generic knowledge of LLMs themselves via prompt engineering while the other is to adopt domain-specific learnable models. Moreover, we introduce a local memory mechanism for achieving the third capability. We evaluate our proposed ResponsibleTA on UI task automation and hope it could bring more attentions to ensuring LLMs more responsible in diverse scenarios. The research project homepage is at https://task-automation-research.github.io/responsible_task_automation.

We release Terminal Wrench, a subset of 331 terminal-agent benchmark environments, copied from the popular open benchmarks that are demonstrably reward-hackable. The data set includes 3,632 hack trajectories and 2,352 legitimate baseline trajectories across three frontier models (Claude Opus 4.6, Gemini 3.1 Pro, GPT-5.4). Each entry preserves the original task definition alongside full attack trajectories that show how the verifier was bypassed. It also includes cases where the task was not solved as intended. The tasks span system administration, machine learning, software engineering, and security challenges; the exploits range from simple output spoofing to stack-frame introspection, standard-library patching, and rootkit-style binary hijacking. Crucially, these exploits are specific to each task, rather than the evaluation harness, making them harder to patch. We also present a monitorability study in which hack trajectories are sanitized or stripped of reasoning traces and then scored by an LLM judge, showing that detection degrades meaningfully when chain-of-thought is removed (AUC drops from 0.97 to 0.92). The data set is publicly available at https://github.com/few-sh/terminal-wrench.

Current benchmarks for computer-use agents evaluate models in impersonal environments. This leaves a gap between evaluation and deployment where personal assistants are expected to work across a user's whole digital life, including their context, historical data, and logged-in accounts. This gap is widest on web tasks, where live web evaluations cannot exercise sites that require logging in or personal information, the kind of site a real personal assistant has to drive. We introduce MyPCBench, which tests computer-use agents as personal assistants on a Linux desktop populated with 17 simulated real-world web applications and a full desktop stack, all seeded for one canonical persona, Michael Scott from The Office. We define 184 tasks in this environment, each inspired by a real request drawn from the OpenClaw community, and benchmark six closed and open-weight models with a uniform computer+bash tool surface. We find that the best model, Claude Opus 4.6, fully solves 55.4\% of the tasks, the only model above 50\%. Model failures cluster on tasks that span many applications and on long trajectories, where personalization stresses an assistant the most. We release the environment, task set, and agent harness at https://mypcbench.com.

Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest achieves 65.36% pass@1, but only 25.25% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox

Efficient task scheduling is paramount in the Linux kernel, where the Completely Fair Scheduler (CFS) meticulously manages CPU resources to balance high utilization with interactive responsiveness. This research pioneers the use of deep learning techniques to predict the sequence of tasks selected by CFS, aiming to evaluate the feasibility of a more generalized and potentially more adaptive task scheduler for diverse workloads. Our core contributions are twofold: first, the systematic generation and curation of a novel scheduling dataset from a running Linux kernel, capturing real-world CFS behavior; and second, the development, training, and evaluation of a Long Short-Term Memory (LSTM) network designed to accurately forecast the next task to be scheduled. This paper further discusses the practical pathways and implications of integrating such a predictive model into the kernel's scheduling framework. The findings and methodologies presented herein open avenues for data-driven advancements in kernel scheduling, with the full source code provided for reproducibility and further exploration.

WANDR (Wide ANd Deep Research) is a benchmark of 500 realistic, challenging data-collection tasks for research agents. Each task requires a system to discover a large set of entities that satisfy specified criteria (breadth), investigate each entity through multiple coordinated web searches (depth), and return independently verifiable records with supporting sources and excerpts. Tasks are represented as qualification key hierarchies that specify the entities, relationships, evidence, and required count at each level; a hierarchy with n companies, m employees per company, and k sources per employee requires n x m x k records. This structure supports diverse workflows such as market mapping, due diligence, literature review, product comparison, and talent sourcing, with targets ranging from dozens to thousands of records. WANDR replaces static gold answer sets with task-specific judges that refetch cited pages and verify each record against its evidence, allowing evaluation of current and changing facts. Record verdicts are aggregated into soft and hard precision, recall, and F1 scores that distinguish factual quality, coverage, and hierarchical completeness. The tasks are derived from de-identified product-usage logs and produced through a semi-automated pipeline with automated checks, empirical audits, and human review where needed. We evaluate six production research systems and find that the benchmark is far from saturated: at high effort, the strongest system reaches only 0.363 soft F1 and 0.133 hard F1. Performance degrades as target volume and hierarchy depth increase, with incomplete discovery, missing enrichment, and incomplete evidence construction remaining major bottlenecks. The benchmark and evaluation harness are available at https://github.com/perplexityai/wandr.

We introduce Codex, a GPT language model fine-tuned on publicly available code from GitHub, and study its Python code-writing capabilities. A distinct production version of Codex powers GitHub Copilot. On HumanEval, a new evaluation set we release to measure functional correctness for synthesizing programs from docstrings, our model solves 28.8% of the problems, while GPT-3 solves 0% and GPT-J solves 11.4%. Furthermore, we find that repeated sampling from the model is a surprisingly effective strategy for producing working solutions to difficult prompts. Using this method, we solve 70.2% of our problems with 100 samples per problem. Careful investigation of our model reveals its limitations, including difficulty with docstrings describing long chains of operations and with binding operations to variables. Finally, we discuss the potential broader impacts of deploying powerful code generation technologies, covering safety, security, and economics.

Can LLM agents make contributions to frontier computer science R&D? We introduce FrontierCS 2, a benchmark of 48 open-ended tasks. We take a paper and the code its authors released and delete the paper's contribution. The deleted code becomes the human reference that sets the bar. Earlier open-ended benchmarks give each agent a problem statement and grade one program after the run. FrontierCS 2 places the agent inside a research environment. The agent inherits a working repository and may submit a patch at any point. For each patch, a judge returns a development-suite score but keeps the evaluator, the held-out final suite and the reference private, and grades the final submission on that suite against the reference. The tasks span 13 domains, from machine-learning systems and high-performance computing to databases, security and robotics. Beyond these 48 tasks, FrontierCS 2 keeps a private set of more than 100 tasks, to be announced in the future. We evaluate 12 agents. The best one is Claude Opus 5.5, passing 58% of its runs. No run passes on 16 tasks. A higher score is not the same as a pass: most runs that beat the reference on the combined final score still fall short on some metric or workload. On a shared capability scale for agents and tasks, the median task is roughly as difficult as what the strongest agent can handle.

We study whether phone-use agents respect privacy while completing benign mobile tasks. This question has remained hard to answer because privacy-compliant behavior is not operationalized for phone-use agents, and ordinary apps do not reveal exactly what data agents type into which form entries during execution. To make this question measurable, we introduce MyPhoneBench, a verifiable evaluation framework for privacy behavior in mobile agents. We operationalize privacy-respecting phone use as permissioned access, minimal disclosure, and user-controlled memory through a minimal privacy contract, iMy, and pair it with instrumented mock apps plus rule-based auditing that make unnecessary permission requests, deceptive re-disclosure, and unnecessary form filling observable and reproducible. Across five frontier models on 10 mobile apps and 300 tasks, we find that task success, privacy-compliant task completion, and later-session use of saved preferences are distinct capabilities, and no single model dominates all three. Evaluating success and privacy jointly reshuffles the model ordering relative to either metric alone. The most persistent failure mode across models is simple data minimization: agents still fill optional personal entries that the task does not require. These results show that privacy failures arise from over-helpful execution of benign tasks, and that success-only evaluation overestimates the deployment readiness of current phone-use agents. All code, mock apps, and agent trajectories are publicly available at~ https://github.com/tangzhy/MyPhoneBench.

We present UI-Venus, a native UI agent that takes only screenshots as input based on a multimodal large language model. UI-Venus achieves SOTA performance on both UI grounding and navigation tasks using only several hundred thousand high-quality training samples through reinforcement finetune (RFT) based on Qwen2.5-VL. Specifically, the 7B and 72B variants of UI-Venus obtain 94.1% / 50.8% and 95.3% / 61.9% on the standard grounding benchmarks, i.e., Screenspot-V2 / Pro, surpassing the previous SOTA baselines including open-source GTA1 and closed-source UI-TARS-1.5. To show UI-Venus's summary and planing ability, we also evaluate it on the AndroidWorld, an online UI navigation arena, on which our 7B and 72B variants achieve 49.1% and 65.9% success rate, also beating existing models. To achieve this, we introduce carefully designed reward functions for both UI grounding and navigation tasks and corresponding efficient data cleaning strategies. To further boost navigation performance, we propose Self-Evolving Trajectory History Alignment & Sparse Action Enhancement that refine historical reasoning traces and balances the distribution of sparse but critical actions, leading to more coherent planning and better generalization in complex UI tasks. Our contributions include the publish of SOTA open-source UI agents, comprehensive data cleaning protocols and a novel self-evolving framework for improving navigation performance, which encourage further research and development in the community. Code is available at https://github.com/inclusionAI/UI-Venus.

The task of issue resolving is to modify a codebase to generate a patch that addresses a given issue. However, existing benchmarks, such as SWE-bench, focus almost exclusively on Python, making them insufficient for evaluating Large Language Models (LLMs) across diverse software ecosystems. To address this, we introduce a multilingual issue-resolving benchmark, called Multi-SWE-bench, covering Java, TypeScript, JavaScript, Go, Rust, C, and C++. It includes a total of 1,632 high-quality instances, which were carefully annotated from 2,456 candidates by 68 expert annotators, ensuring that the benchmark can provide an accurate and reliable evaluation. Based on Multi-SWE-bench, we evaluate a series of state-of-the-art models using three representative methods (Agentless, SWE-agent, and OpenHands) and present a comprehensive analysis with key empirical insights. In addition, we launch a Multi-SWE-RL open-source community, aimed at building large-scale reinforcement learning (RL) training datasets for issue-resolving tasks. As an initial contribution, we release a set of 4,723 well-structured instances spanning seven programming languages, laying a solid foundation for RL research in this domain. More importantly, we open-source our entire data production pipeline, along with detailed tutorials, encouraging the open-source community to continuously contribute and expand the dataset. We envision our Multi-SWE-bench and the ever-growing Multi-SWE-RL community as catalysts for advancing RL toward its full potential, bringing us one step closer to the dawn of AGI.

Autonomous agents are increasingly adopted to complete complex, multi-tool workflows in real-world settings. However, existing benchmarks typically separate tasks by application or capability and evaluate agents in environments that are cleaner and more stable than those encountered in practice. We introduce DuMateBench, a real-session benchmark reconstructed from anonymized and privacy-screened user sessions collected from a large-scale production agent platform. Each task preserves the relevant pre-solution interaction history, persistent configurations, and workspace state, and is then validated through human verification. The resulting benchmark comprises 200 tasks spanning 8 broad scenarios and 17 fine-grained capability categories, with most tasks requiring multiple capability coordination. We execute these tasks in isolated Docker containers injected with three forms of real-world environmental complexity: Insufficient, Unstable, and Noisy, and assess performance using a hybrid deterministic and LLM-as-Judge evaluation protocol. Experiments across five representative autonomous-agent frameworks paired with four state-of-the-art LLMs reveal substantial gaps in strict task completion. Complementary robustness, efficiency, and diagnostic analyses further show that performance under environmental perturbations is jointly shaped by the capabilities of the LLM and the surrounding agent framework. The code and data are publicly available at https://dumatebench.com/.

The rise of convenience packaging has led to generation of enormous waste, making efficient waste sorting crucial for sustainable waste management. To address this, we developed DWaste, a computer vision-powered platform designed for real-time waste sorting on resource-constrained smartphones and edge devices, including offline functionality. We benchmarked various image classification models (EfficientNetV2S/M, ResNet50/101, MobileNet) and object detection (YOLOv8n, YOLOv11n) using a subset of our own waste data set and annotated it using the custom tool Annotated Lab. We found a clear trade-off between accuracy and resource consumption: the best classifier, EfficientNetV2S, achieved high accuracy (~ 96%) but suffered from high latency (~ 0.22s) and elevated carbon emissions. In contrast, lightweight object detection models delivered strong performance (up to 77% mAP) with ultra-fast inference (~ 0.03s) and significantly smaller model sizes (< 7MB), making them ideal for real-time, low-power use. Model quantization further maximized efficiency, substantially reducing model size and VRAM usage by up to 75%. Our work demonstrates the successful implementation of "Greener AI" models to support real-time, sustainable waste sorting on edge devices.

Agentic coding requires agents to effectively interact with runtime environments, e.g., command line interfaces (CLI), so as to complete tasks like resolving dependency issues, fixing system problems, etc. But it remains underexplored how such environment-intensive tasks can be obtained at scale to enhance agents' capabilities. To address this, based on an analogy between the Dockerfile and the agentic task, we propose to employ agents to simulate and explore environment histories, guided by execution feedback. By tracing histories of a healthy environment, its state can be inverted to an earlier one with runtime failures, from which a task can be derived by packing the buggy state and the corresponding error messages. With our method, named CLI-Gym, a total of 1,655 environment-intensive tasks are derived, being the largest collection of its kind. Moreover, with curated successful trajectories, our fine-tuned model, named LiberCoder, achieves substantial absolute improvements of +21.1% (to 46.1%) on Terminal-Bench, outperforming various strong baselines. To our knowledge, this is the first public pipeline for scalable derivation of environment-intensive tasks.

Training diffusion models with multiple rewards requires distinguishing user preference from reward informativeness. User preference determines how much each reward should contribute to the overall objective; reward informativeness determines when its feedback is useful during denoising. Some rewards can meaningfully evaluate a sample as soon as global structure emerges, but others become informative only when the sample is nearly clean. To address both questions jointly, we propose ReCAST (Reward Credit ASsignment across T}imesteps), the first method, to our knowledge, for per-reward, timestep-dependent credit assignment in diffusion reward fine-tuning. ReCAST separates user preferences from temporal allocation through a reward-by-timestep weight matrix W, whose row sums match the user-specified reward budgets λ, while its column sums are equal, assigning the same total weight to each denoising step. Under these marginal constraints, ReCAST allocates weight according to each reward's informativeness, quantified by its Rényi discriminability gain at each step. These gains telescope to the total discriminability between the reward-induced positive policy and the current policy, providing a basis for temporal credit assignment. We evaluate ReCAST by training SD3.5-Medium under two distinct four-reward settings, each across five reward budgets λ. ReCAST improves the training rewards in one setting and matches them in the other, improves every held-out judge in both, and is preferred by an independent LLM-as-a-Judge. Together, these results show that ReCAST yields improvements that generalize beyond the training rewards and support its core principle: assigning each reward greater weight at the denoising timesteps where its feedback is most informative.

One of the challenges in applying reinforcement learning in a complex real-world environment lies in providing the agent with a sufficiently detailed reward function. Any misalignment between the reward and the desired behavior can result in unwanted outcomes. This may lead to issues like "reward hacking" where the agent maximizes rewards by unintended behavior. In this work, we propose to disentangle the reward into two distinct parts. A simple task-specific reward, outlining the particulars of the task at hand, and an unknown common-sense reward, indicating the expected behavior of the agent within the environment. We then explore how this common-sense reward can be learned from expert demonstrations. We first show that inverse reinforcement learning, even when it succeeds in training an agent, does not learn a useful reward function. That is, training a new agent with the learned reward does not impair the desired behaviors. We then demonstrate that this problem can be solved by training simultaneously on multiple tasks. That is, multi-task inverse reinforcement learning can be applied to learn a useful reward function.

The issue-resolving task, where a model generates patches to fix real-world bugs, has emerged as a critical benchmark for evaluating the capabilities of large language models (LLMs). While SWE-bench and its variants have become standard in this domain, they suffer from key limitations: they have not been updated since their initial releases, cover a narrow set of repositories, and depend heavily on manual effort for instance construction and environment setup. These factors hinder scalability and introduce risks of overfitting and data contamination. In this work, we present SWE-bench-Live, a live-updatable benchmark designed to overcome these challenges. Our initial release consists of 1,319 tasks derived from real GitHub issues created since 2024, spanning 93 repositories. Each task is accompanied by a dedicated Docker image to ensure reproducible execution. Central to our benchmark is \method, an automated curation pipeline that streamlines the entire process from instance creation to environment setup, removing manual bottlenecks and enabling scalability and continuous updates. We evaluate a range of state-of-the-art agent frameworks and LLMs on SWE-bench-Live, revealing a substantial performance gap compared to static benchmarks like SWE-bench, even under controlled evaluation conditions. To better understand this discrepancy, we perform detailed analyses across repository origin, issue recency, and task difficulty. By providing a fresh, diverse, and executable benchmark grounded in live repository activity, SWE-bench-Live facilitates rigorous, contamination-resistant evaluation of LLMs and agents in dynamic, real-world software development settings.

For a robot to personalize physical assistance effectively, it must learn user preferences that can be generally reapplied to future scenarios. In this work, we investigate personalization of household cleanup with robots that can tidy up rooms by picking up objects and putting them away. A key challenge is determining the proper place to put each object, as people's preferences can vary greatly depending on personal taste or cultural background. For instance, one person may prefer storing shirts in the drawer, while another may prefer them on the shelf. We aim to build systems that can learn such preferences from just a handful of examples via prior interactions with a particular person. We show that robots can combine language-based planning and perception with the few-shot summarization capabilities of large language models (LLMs) to infer generalized user preferences that are broadly applicable to future interactions. This approach enables fast adaptation and achieves 91.2% accuracy on unseen objects in our benchmark dataset. We also demonstrate our approach on a real-world mobile manipulator called TidyBot, which successfully puts away 85.0% of objects in real-world test scenarios.

Large language model (LLM) agents have demonstrated remarkable capabilities in complex decision-making and tool-use tasks, yet their ability to generalize across varying environments remains a under-examined concern. Current evaluation paradigms predominantly rely on trajectory-based metrics that measure task success, while failing to assess whether agents possess a grounded, transferable model of the environment. To address this gap, we propose Task-to-Quiz (T2Q), a deterministic and automated evaluation paradigm designed to decouple task execution from world-state understanding. We instantiate this paradigm in T2QBench, a suite comprising 30 environments and 1,967 grounded QA pairs across multiple difficulty levels. Our extensive experiments reveal that task success is often a poor proxy for environment understanding, and that current memory machanism can not effectively help agents acquire a grounded model of the environment. These findings identify proactive exploration and fine-grained state representation as primary bottlenecks, offering a robust foundation for developing more generalizable autonomous agents.

As data-science agents shift from co-pilots to auto-pilots, silent misframing becomes a critical failure mode. Agents quietly commit to plausible but unintended task framings, producing clean, executable artifacts that hide their incorrect assessment of the task. Existing benchmarks score whether the pipeline runs, ignoring whether the agent recognized the task was underspecified. We introduce Ambig-DS, two diagnostic suites: one for prediction-target ambiguity (Ambig-DS-Target, 51 tasks built on DSBench, a tabular modeling benchmark) and one for evaluation-objective ambiguity (Ambig-DS-Objective, 61 tasks built on MLE-bench, a Kaggle-style ML competition benchmark), constructed so that scoring uses each source benchmark's original evaluator. For every task we pair the original, fully specified version with an ambiguous variant produced by controlled edits; a human-and-LLM verification pipeline confirms each variant admits multiple plausible interpretations with decision-relevant consequences. The suites are analyzed independently and ambiguity lowers performance in both. Across five agents spanning efficient to frontier-class models, we find in our controlled diagnostic setting: (i) failures are silent commitments: wrong-target submissions on Target, wrong-metric or non-committal baseline submissions on Objective, rather than execution errors; (ii) allowing the agent to ask one clarifying question recovers much of the loss under idealized conditions, suggesting missing framing information drives a substantial part of the observed degradation; but (iii) agents cannot reliably tell when to use it: permissive prompts induce over-asking on clear tasks, while conservative prompts induce silent defaulting on ambiguous ones. Recognizing target and objective underspecification, not pipeline execution, is the bottleneck missing from standard DS-agent evaluations.

We introduce SWE-Bench Pro, a substantially more challenging benchmark that builds upon the best practices of SWE-BENCH [25], but is explicitly designed to capture realistic, complex, enterprise-level problems beyond the scope of SWE-BENCH. SWE-BENCH PRO contains 1,865 problems sourced from a diverse set of 41 actively maintained repositories spanning business applications, B2B services, and developer tools. The benchmark is partitioned into a public set with open access to problems sourced from 11 repositories, a held-out set of 12 repositories and a commercial set of 18 proprietary repositories where we have formal partnership agreements with early-stage startups. Problems in the held-out and the commercial set are not publicly accessible, but we release results on the commercial set. Our benchmark features long-horizon tasks that may require hours to days for a professional software engineer to complete, often involving patches across multiple files and substantial code modifications. All tasks are human-verified and augmented with sufficient context to ensure resolvability. In our evaluation of widely used coding models, under a unified scaffold, we observe that their performance on SWE-Bench PRO remains below 25% (Pass@1), with GPT-5 achieving the highest score to date at 23.3%. To better understand these limitations, we cluster the failure modes observed in the collected agent trajectories for a clearer characterization of the error patterns exhibited by current models. Overall, SWE-BENCH PRO provides a contamination-resistant testbed that more faithfully captures the complexity and diversity of real-world software development, advancing the pursuit of truly autonomous software engineering agents at a professional level.

Every home is different, and every person likes things done in their particular way. Therefore, home robots of the future need to both reason about the sequential nature of day-to-day tasks and generalize to user's preferences. To this end, we propose a Transformer Task Planner(TTP) that learns high-level actions from demonstrations by leveraging object attribute-based representations. TTP can be pre-trained on multiple preferences and shows generalization to unseen preferences using a single demonstration as a prompt in a simulated dishwasher loading task. Further, we demonstrate real-world dish rearrangement using TTP with a Franka Panda robotic arm, prompted using a single human demonstration.

We present a straightforward statistical test to detect certain violations of the assumption that the data are Independent and Identically Distributed (IID). The specific form of violation considered is common across real-world applications: whether the examples are ordered in the dataset such that almost adjacent examples tend to have more similar feature values (e.g. due to distributional drift, or attractive interactions between datapoints). Based on a k-Nearest Neighbors estimate, our approach can be used to audit any multivariate numeric data as well as other data types (image, text, audio, etc.) that can be numerically represented, perhaps with model embeddings. Compared with existing methods to detect drift or auto-correlation, our approach is both applicable to more types of data and also able to detect a wider variety of IID violations in practice. Code: https://github.com/cleanlab/cleanlab

Large language and vision-language models increasingly power agents that act on a user's behalf through command-line interface (CLI) harnesses. However, most agent benchmarks still rely on synthetic sandboxes, short-horizon tasks, mock-service APIs, and final-answer checks, leaving open whether agents can complete realistic long-horizon work in the runtimes where they are deployed. This work presents WildClawBench, a native-runtime benchmark of 60 human-authored, bilingual, multimodal tasks spanning six thematic categories. Each task averages roughly 8 minutes of wall-clock time and over 20 tool calls, and runs inside a reproducible Docker container hosting an actual CLI agent harness (OpenClaw, Claude Code, Codex, or Hermes Agent) with access to real tools rather than mock services. Grading is hybrid, combining deterministic rule-based checks, environment-state auditing of side effects, and an LLM/VLM judge for semantic verification. Across 19 frontier models, the best, Claude Opus 4.7, reaches only 62.2% overall under OpenClaw, while every other model stays below 60%, and switching harness alone shifts a single model by up to 18 points. These results show that long-horizon, native-runtime agent evaluation remains a far-from-resolved task for current frontier models. We release the tasks, code, and containerized tooling to support reproducible evaluation.

A declared task can sometimes certify a reduced source before a downstream encoder, codebook, rate, distortion target, or optimizer is chosen. This paper studies when one such reduction preserves the complete downstream problem family, a property termed Task-Sufficient Contraction. The reduced source is fixed by the task before the later operating point is selected. An exact contraction allows the later problem to be solved on that source with the same result as if the full source had been retained. For a machine with a fixed set of possible actions and a fixed loss, the paper identifies a consumer-specific source by merging states only when every available action has the same regret in both. For finite action sets, replacing the richer source by this reduced source preserves the complete one-step rate-regret curve, even though the reduction is fixed before the distortion target is chosen. A second result gives an exact characterization for quadratic loss on affine feasible-action sets: the canonical reduced source is the projection onto the directions in which feasible actions can differ. Under a fixed energy budget, this becomes centered load, while retaining only the optimal water-filled action is too coarse. Earlier Information Bottleneck, semantic rate-distortion, and goal-oriented quantization results are then used to distinguish exact, architecture-conditioned, approximate, failed, and corrected contractions. The framework suggests a way for heterogeneous machines to exchange what a receiving task needs without first aligning their full internal representations.

In this paper, we introduce OneReward, a unified reinforcement learning framework that enhances the model's generative capabilities across multiple tasks under different evaluation criteria using only One Reward model. By employing a single vision-language model (VLM) as the generative reward model, which can distinguish the winner and loser for a given task and a given evaluation criterion, it can be effectively applied to multi-task generation models, particularly in contexts with varied data and diverse task objectives. We utilize OneReward for mask-guided image generation, which can be further divided into several sub-tasks such as image fill, image extend, object removal, and text rendering, involving a binary mask as the edit area. Although these domain-specific tasks share same conditioning paradigm, they differ significantly in underlying data distributions and evaluation metrics. Existing methods often rely on task-specific supervised fine-tuning (SFT), which limits generalization and training efficiency. Building on OneReward, we develop Seedream 3.0 Fill, a mask-guided generation model trained via multi-task reinforcement learning directly on a pre-trained base model, eliminating the need for task-specific SFT. Experimental results demonstrate that our unified edit model consistently outperforms both commercial and open-source competitors, such as Ideogram, Adobe Photoshop, and FLUX Fill [Pro], across multiple evaluation dimensions. Code and model are available at: https://one-reward.github.io

We present MineNPC-Task, a user-authored benchmark and evaluation harness for testing memory-aware, mixed-initiative LLM agents in open-world Minecraft. Rather than relying on synthetic prompts, tasks are elicited through formative and summative co-play with expert players, then normalized into parametric templates with explicit preconditions and dependency structure. These tasks are paired with machine-checkable validators under a bounded-knowledge policy that forbids out-of-world shortcuts. The harness captures plan, action, and memory events, including plan previews, targeted clarifications, memory reads and writes, precondition checks, and repair attempts, and reports outcomes relative to the total number of attempted subtasks using only in-world evidence. As an initial snapshot, we instantiate the framework with GPT-4o and evaluate 216 subtasks across 8 experienced players. We observe recurring breakdown patterns in code execution, inventory and tool handling, referencing, and navigation, alongside successful recoveries supported by mixed-initiative clarifications and lightweight memory use. Participants rated interaction quality and interface usability positively, while noting the need for stronger memory persistence across tasks. We release the complete task suite, validators, logs, and evaluation harness to support transparent and reproducible evaluation of future memory-aware embodied agents.

Despite rapid progress, generalist robot policies remain brittle on complex, long-horizon tasks that comprise multiple stages or require repeated attempts and deliberation on the same underlying stage before success. Q-value functions can improve these policies by ranking candidate actions or guiding policy improvement, but learning from sparse task-level rewards entails long credit-assignment horizons, difficult Bellman backups, and broad data-coverage requirements. We introduce SeeQ (Subtask-elicited Q-functions), which instead learns Q-values for the currently active subtask. This shortens the value-prediction horizon and enables effective learning with temporal-difference (TD) objectives. During training, subtask-level annotations present in offline robot data provide the decomposition and enable learning from broad, potentially suboptimal robot datasets. To eliminate the need for human annotations or modular subtask prediction systems at test time, our Q-function architecture is trained to autoregressively predict the active subtask in natural language before estimating its value. We instantiate SeeQ using a base vision-language backbone, pretrain it on diverse open-source robot manipulation data, and finetune it on downstream tasks. Across four real-world manipulation tasks on two bimanual robot platforms, the SeeQ value function substantially improves best-of-N policy steering.

Modern web interfaces are unnecessarily complex to use as they overwhelm users with excessive text and visuals unrelated to their current goals. This problem particularly impacts screen reader users (SRUs), who navigate content sequentially and may spend minutes traversing irrelevant elements before reaching desired information compared to vision users (VUs) who visually skim in seconds. We present Task Mode, a system that dynamically filters web content based on user-specified goals using large language models to identify and prioritize relevant elements while minimizing distractions. Our approach preserves page structure while offering multiple viewing modes tailored to different access needs. Our user study with 12 participants (6 VUs, 6 SRUs) demonstrates that our approach reduced task completion time for SRUs while maintaining performance for VUs, decreasing the completion time gap between groups from 2x to 1.2x. 11 of 12 participants wanted to use Task Mode in the future, reporting that Task Mode supported completing tasks with less effort and fewer distractions. This work demonstrates how designing new interactions simultaneously for visual and non-visual access can reduce rather than reinforce accessibility disparities in future technology created by human-computer interaction researchers and practitioners.

English-only benchmarks expose only a narrow slice of voice-agent behavior. We introduce τ-Multilingual, extending τ-Voice to Spanish, Brazilian Portuguese, Hindi, Korean, and Mandarin with native-speaker review and evaluation of generated language and spoken output. Across 4,500 full-duplex calls and five voice configurations, Spanish, Portuguese, and Hindi remain within 3.2 task-completion points of English, but Korean and Mandarin fall by 14.7 and 8.4 points. The failure modes also vary: Korean systems miss more responses, Mandarin systems interrupt more often, and both struggle with tools and entities. Grok leads task completion but scores lowest on generation quality, motivating separate task, interaction, and generation reporting. We release language packs, validated judges, and tools for community-built multilingual voice-agent evaluation.

Coding agents repeatedly consume long tool observations even though only a small fraction of each observation matters for the next step. We study task-conditioned tool-output pruning: given a focused query and one tool output, return the smallest verbatim evidence block the agent should inspect next. We introduce a benchmark of 11,477 examples built from SWE-bench repository interactions and synthetic multi-ecosystem tool outputs, with a manually curated 618-example test set. We fine-tune Qwen 3.5 2B with LoRA and compare it against larger zero-shot models and heuristic pruning baselines. Our model reaches 0.86 recall and 0.80 F1 while removing 92% of input tokens, outperforming zero-shot Qwen 3.5 35B A3B by 11 recall points and all heuristic baselines by a wide margin.

Robot models, particularly those trained with large amounts of data, have recently shown a plethora of real-world manipulation and navigation capabilities. Several independent efforts have shown that given sufficient training data in an environment, robot policies can generalize to demonstrated variations in that environment. However, needing to finetune robot models to every new environment stands in stark contrast to models in language or vision that can be deployed zero-shot for open-world problems. In this work, we present Robot Utility Models (RUMs), a framework for training and deploying zero-shot robot policies that can directly generalize to new environments without any finetuning. To create RUMs efficiently, we develop new tools to quickly collect data for mobile manipulation tasks, integrate such data into a policy with multi-modal imitation learning, and deploy policies on-device on Hello Robot Stretch, a cheap commodity robot, with an external mLLM verifier for retrying. We train five such utility models for opening cabinet doors, opening drawers, picking up napkins, picking up paper bags, and reorienting fallen objects. Our system, on average, achieves 90% success rate in unseen, novel environments interacting with unseen objects. Moreover, the utility models can also succeed in different robot and camera set-ups with no further data, training, or fine-tuning. Primary among our lessons are the importance of training data over training algorithm and policy class, guidance about data scaling, necessity for diverse yet high-quality demonstrations, and a recipe for robot introspection and retrying to improve performance on individual environments. Our code, data, models, hardware designs, as well as our experiment and deployment videos are open sourced and can be found on our project website: https://robotutilitymodels.com

When trying to solve a computational problem, we are often faced with a choice between algorithms that are guaranteed to return the right answer but differ in their runtime distributions (e.g., SAT solvers, sorting algorithms). This paper aims to lay theoretical foundations for such choices by formalizing preferences over runtime distributions. It might seem that we should simply prefer the algorithm that minimizes expected runtime. However, such preferences would be driven by exactly how slow our algorithm is on bad inputs, whereas in practice we are typically willing to cut off occasional, sufficiently long runs before they finish. We propose a principled alternative, taking a utility-theoretic approach to characterize the scoring functions that describe preferences over algorithms. These functions depend on the way our value for solving our problem decreases with time and on the distribution from which captimes are drawn. We describe examples of realistic utility functions and show how to leverage a maximum-entropy approach for modeling underspecified captime distributions. Finally, we show how to efficiently estimate an algorithm's expected utility from runtime samples.

We introduce Dynatask: an open source system for setting up custom NLP tasks that aims to greatly lower the technical knowledge and effort required for hosting and evaluating state-of-the-art NLP models, as well as for conducting model in the loop data collection with crowdworkers. Dynatask is integrated with Dynabench, a research platform for rethinking benchmarking in AI that facilitates human and model in the loop data collection and evaluation. To create a task, users only need to write a short task configuration file from which the relevant web interfaces and model hosting infrastructure are automatically generated. The system is available at https://dynabench.org/ and the full library can be found at https://github.com/facebookresearch/dynabench.

Like generic multi-task learning, continual learning has the nature of multi-objective optimization, and therefore faces a trade-off between the performance of different tasks. That is, to optimize for the current task distribution, it may need to compromise performance on some previous tasks. This means that there exist multiple models that are Pareto-optimal at different times, each addressing a distinct task performance trade-off. Researchers have discussed how to train particular models to address specific trade-off preferences. However, existing algorithms require training overheads proportional to the number of preferences -- a large burden when there are multiple, possibly infinitely many, preferences. As a response, we propose Imprecise Bayesian Continual Learning (IBCL). Upon a new task, IBCL (1) updates a knowledge base in the form of a convex hull of model parameter distributions and (2) obtains particular models to address task trade-off preferences with zero-shot. That is, IBCL does not require any additional training overhead to generate preference-addressing models from its knowledge base. We show that models obtained by IBCL have guarantees in identifying the Pareto optimal parameters. Moreover, experiments on standard image classification and NLP tasks support this guarantee. Statistically, IBCL improves average per-task accuracy by at most 23% and peak per-task accuracy by at most 15% with respect to the baseline methods, with steadily near-zero or positive backward transfer. Most importantly, IBCL significantly reduces the training overhead from training 1 model per preference to at most 3 models for all preferences.

Coding agents such as Claude Code also follow text files that developers write for them: a project file, skills the agent may choose to open, and rules that load automatically when the agent reads matching files. We call these files the agent's harness. Developers add to it whenever they correct the agent, but one success proves little: the same agent can pass a task once and fail it the next time. We present Cortex, a tool that tests changes to the harness instead of trusting them. It saves fixes from the developer's sessions as replayable tasks, finds the corrections that recur, writes one skill or rule at a time, and keeps it only if repeated runs with and without it show a net gain and no task made noticeably worse. Existing text is tested the same way before it is deleted, and Cortex neither adds nor deletes text the agent never used. We tested Cortex with Claude Haiku 4.5 on a repository built for the purpose, with conventions the model tends to break, one area with nothing to learn, and a script playing the developer. On held-out tasks, the harnesses it built made the agent follow the conventions far more often than no skills or rules did, left the comparison area unchanged, and added no text read on every turn. Rules, which load automatically, did most of the work; skills helped only when offered with the files a task touches, and those offered on every turn were rarely opened. Cortex kept none of the useless or harmful suggestions we planted, but the agent never opened the useless ones or the harmful skills, and it reached no verdict on the harmful rules, so the test shows that Cortex keeps no unused text, not that it detects damage. Under the stronger Claude Sonnet 5 a single run's harness added little, and on a real open-source project, where the agent was rarely corrected, Cortex found nothing to learn. We release the tool, the lab and its data.

Scaling computer-use agents from individual actions to complex software work requires experience that connects decisions to complete outcomes. We present NorthCUA, an integrated system for turbocharging computer-use agents through experience at scale. Its Experience Flywheel supplies diverse executable tasks to the NorthCUA Harness, which sustains long interactions, records decisions, and checks task outcomes. The Flywheel uses these records to curate verified trajectories. For unsolved tasks, recursive self-improvement evolves reusable skills and preserves verified milestones so later attempts build on earlier progress. Staged Learning combines this experience with grounding data from the Flywheel to develop GUI competence through mid-training and long-horizon supervised fine-tuning. The policy then learns from fresh attempts in the same Harness through joint desktop-web reinforcement learning, closing the loop between experience production and policy improvement. The resulting Nano, Mini, and Pro family achieves state-of-the-art results among open-source models across GUI grounding, desktop execution, and Web interaction under the reported evaluation settings. For more responsive Web workflows, NorthCUA-Ultrafast pairs direct action selection with open-ended generation on pre-RL Nano, offering a measured quality-latency tradeoff. Together, these advances establish NorthCUA as a high-performing model family and a reusable foundation for agents that carry complex software work from intent to completion.

AI agents increasingly carry out long-horizon professional work, but their evaluations rarely require a finished creative deliverable. To this end, we introduce Timeline-Bench, a benchmark of 56 real video-editing tasks, each asking an agent to turn raw production material into a finished video. Tasks range from selecting dialog takes and shaping interview footage into a story to cutting commercials from product shots, voiceovers and graphics. Every task provides a brief, source assets, a container and a set of tests. A task is resolved when the output passes every test. The tests check the delivery format, the content and the brief's explicit requirements, and include a quality test calibrated on 2,582 blind judgments by 43 video editors. We evaluate 16 agents that pair frontier models with coding-agent harnesses such as Codex, Claude Code and OpenCode. The best, GPT-6 Astra in Codex with curated editorial guidance, resolves only 15 of the 56 tasks (26.8%), and the average agent resolves 14.0%. Human editors prefer the reference edit in 83.5% of judgments. Most unresolved runs (562 of 771) fail only the quality test: agents perceive footage through stills and transcripts and check their renders for defects, not craft. We release the tasks, verifier and per-run results at https://timelinebench.tensortest.com.

Much knowledge work produces new deliverables from files a workspace already holds, and LLM agents are beginning to take such work over. Through direct corpus interaction, an agent can search and read any of those files from a terminal with no indexing, and producing a deliverable from many of them in this way is what we call direct workspace interaction (DWI). Reaching the files, however, is only half the task: nothing keeps track of what the task asks for, what has been read, and what was listed but never opened, all of which slip through the context window without leaving a trace, so an agent may extract a figure and still deliver a report without it. To address this, we present RunningTab, a framework that equips direct workspace interaction with an environment-side tab: a per-task record of what the task still owes, kept by the environment alongside the agent. Specifically, the agent adds its requirements, while the environment records every file read as an excerpt with its provenance and every listed but unopened file as a candidate; the agent can then see each requirement beside its best-matching excerpts and top unopened candidates, resolve it against matching content or set it aside with a reason, and, should it try to finish with requirements still open, receive them in a finish check. We validate RunningTab on three benchmarks with three LLMs, where it consistently outperforms plain DWI and baselines that keep the record in the model, while its tab usually holds the values a deliverable needs once seen.

Diffusion Transformer has demonstrated powerful capability and scalability in generating high-quality images and videos. Further pursuing the unification of generation and editing tasks has yielded significant progress in the domain of image content creation. However, due to the intrinsic demands for consistency across both temporal and spatial dynamics, achieving a unified approach for video synthesis remains challenging. We introduce VACE, which enables users to perform Video tasks within an All-in-one framework for Creation and Editing. These tasks include reference-to-video generation, video-to-video editing, and masked video-to-video editing. Specifically, we effectively integrate the requirements of various tasks by organizing video task inputs, such as editing, reference, and masking, into a unified interface referred to as the Video Condition Unit (VCU). Furthermore, by utilizing a Context Adapter structure, we inject different task concepts into the model using formalized representations of temporal and spatial dimensions, allowing it to handle arbitrary video synthesis tasks flexibly. Extensive experiments demonstrate that the unified model of VACE achieves performance on par with task-specific models across various subtasks. Simultaneously, it enables diverse applications through versatile task combinations. Project page: https://ali-vilab.github.io/VACE-Page/.

This paper focuses on text detoxification, i.e., automatically converting toxic text into non-toxic text. This task contributes to safer and more respectful online communication and can be considered a Text Style Transfer (TST) task, where the text style changes while its content is preserved. We present three approaches: knowledge transfer from a similar task, multi-task learning approach, combining sequence-to-sequence modeling with various toxicity classification tasks, and delete and reconstruct approach. To support our research, we utilize a dataset provided by Dementieva et al.(2021), which contains multiple versions of detoxified texts corresponding to toxic texts. In our experiments, we selected the best variants through expert human annotators, creating a dataset where each toxic sentence is paired with a single, appropriate detoxified version. Additionally, we introduced a small Hindi parallel dataset, aligning with a part of the English dataset, suitable for evaluation purposes. Our results demonstrate that our approach effectively balances text detoxication while preserving the actual content and maintaining fluency.

Large LMs such as GPT-3 are powerful, but can commit mistakes that are obvious to humans. For example, GPT-3 would mistakenly interpret "What word is similar to good?" to mean a homophone, while the user intended a synonym. Our goal is to effectively correct such errors via user interactions with the system but without retraining, which will be prohibitively costly. We pair GPT-3 with a growing memory of recorded cases where the model misunderstood the user's intents, along with user feedback for clarification. Such a memory allows our system to produce enhanced prompts for any new query based on the user feedback for error correction on similar cases in the past. On four tasks (two lexical tasks, two advanced ethical reasoning tasks), we show how a (simulated) user can interactively teach a deployed GPT-3, substantially increasing its accuracy over the queries with different kinds of misunderstandings by the GPT-3. Our approach is a step towards the low-cost utility enhancement for very large pre-trained LMs. Code, data, and instructions to implement MEMPROMPT for a new task at https://www.memprompt.com/.

This paper reports a single, fully instrumented case study of a large-scale architectural refactoring by an AI coding agent under a specification-first protocol, with no human review of the generated code and no pre-existing oracle to validate the target behaviour. The task, dismantling a central invariant across a large interdependent codebase, was assessed by the author as effectively infeasible through incremental refactoring, the kind of change that conventionally calls for a rewrite instead. Under the protocol described here, the agent completed it successfully. The system is a 717,725-line production TypeScript application across 3,648 files. The task required dismantling a core lifetime invariant: the guarantee that a UI panel remains open for the duration of an AI request. The target behaviour was that a streaming generation survives the closing of its panel and can be reattached, on reopening, to the same live stream with no loss or duplication. The protocol: formal specification by the agent, 14 refinement cycles auditing that specification against the source code, atomic implementation, a compile/test feedback loop, then 17 verification cycles auditing the code against the frozen specification. Across 31 audit passes, 201 defects were corrected before any human executed the program. The convergence criterion was empirical: two consecutive verification passes returning zero findings. The change touched 189 files (31 new); with the extraction phase, the two commits total 288 files, 34,770 insertions, 16,422 deletions. Across the first and roughly thirty later sessions, the software behaved as specified, no bug observed. Elapsed: three days; cost: USD 2,430. The full specification and raw session logs, 1,500+ pages in French, are published as evidence, allowing inspection of the process and submission to a language model for consistency checking.

As executable environments for language agents scale, generating and verifying their rollouts becomes increasingly expensive. Under finite rollout budgets, this raises an important question: can data management at the environment level make agentic reinforcement learning (RL) more efficient by organizing, evaluating, and scheduling task instances together with the environments that determine their tools, states, transitions, and verification? To address this question, we introduce EnvSched (Environment-aware Scheduling), an online scheduler that learns from rollout history. EnvSched assigns scheduling value through two complementary modeling choices: exploitation prioritizes tasks likely to provide informative reward differences, while coverage spreads each batch across underobserved environments and interaction patterns. EnvSched organizes environment--task pairs as nodes in a dual-relation graph, with within-environment edges capturing task semantics and shared tool, state, and verifier structure, and cross-environment edges capturing abstract execution and reward topology. By propagating historical rollout statistics along both edge types, it predicts within-task reward variance. These predictions guide a scheduler that allocates each batch to tasks with high predicted reward variance and underobserved environments and tasks. Environment quotas and a similarity penalty further limit concentration and redundancy, yielding informative and diverse batches without additional pilot rollouts. In RL on Qwen3-8B, EnvSched achieves an average score of 48.81 across BFCL-v3, tau-Bench, and ACEBench-Agent after 40 training steps, exceeding the highest score attained by random task selection after 120 steps. It also improves downstream performance across model sizes at the same training budget.

Multi-task learning over heterogeneous data is fundamental to modern recommendation, while generative models are emerging as the backbone of next-generation recommenders. However, the integration of multi-task learning into the generative paradigm remains largely unexplored. Existing multi-task recommenders, in both discriminative and generative paradigms, extract task-relevant features from a single task-agnostic representation and wire tasks into a predefined conversion funnel. We show that this scheme is inherently prone to a threefold collapse. Source collapse, where task-specific signals are injected late and diluted in the shared latent space. Relational collapse, where task dependencies are either implicitly absorbed by the backbone or statically fixed by predefined funnels. Hierarchical collapse, where tasks depend on features at different scales and shift across training stages. We propose IntHQ, a multi-task generative recommender with three components, each alleviating one collapse. Dual-Stream Decoupling (DSD) injects task identity into computation stream early and separates the shared context stream from the task-specific stream, alleviating signal dilution. Task-Interactive Modeling (TIM) replaces the predefined funnel with explicit cross-task interaction, letting each task condition on the realized outcomes of its predecessors with learned, input-adaptive strength. Hierarchical Querying (HQ) lets each task gather multi-scale information across different layers at different training stages. In offline evaluations, IntHQ consistently outperforms competitive encoder backbones under four representative task-head configurations. Deployed in production on Amap, serving hundreds of millions of users for travel recommendation, IntHQ yields a 1.60\% relative UVCTR lift.

Modern LLM agents are often improved by modifying prompts, tools, or workflows manually, while the executable scaffold surrounding the model---the harness---is typically treated as a fixed artifact after deployment. This work studies an alternative where the harness is task-specific and continuously evolvable: each task family maintains its own harness, which is hot-swapped across iterations through a fixed task-injection seam and rewritten using environment feedback. We introduce Hierarchical Self-Improvement (HSI), a framework in which a single frozen LLM M operates across three hierarchical scopes: a task harness H that executes tasks, an evolver that rewrites H, and a meta-evolver that rewrites the evolver's strategy code under a frozen outer anchor. A thinking-on/off design isolates the contribution of harness evolution by disabling reasoning during task execution while enabling it during self-modification. HSI is bounded by two factors: a feedback-fidelity bound, since evolution requires informative reward signals to guide selection, and a backbone capability bound, since harness redesign cannot overcome limitations of the frozen model. On BALROG with DeepSeek-V4-Flash-Preview as the frozen backbone, HSI achieves consistent gains over the initial harness on moderate-difficulty tasks (+39.3 on BabyAI, +33.0 on Crafter, +25.0 on TextWorld, and +15.0 on MiniHack, all in raw \% Progress), while obtaining strong held-out generalization on BabaIsAI sub-suites (0.98 best-test on BreakStop and 1.00 on GoTo from a 20% unseen split). On tasks beyond the backbone's capability (NLE), harness evolution provides no improvement. These results demonstrate task-specific harness evolution as a viable axis for improving frozen LLM agents under clear empirical limits. Code is available at https://github.com/TailinZhou/hsi.

We present FrogNano, a 4B coding agent designed to tackle software engineering (SWE) tasks efficiently and effectively, even under resource-constrained environments. It is post-trained exclusively via RL on around 1,500 SWE environments with synthetic tasks. A key ingredient for improving performance is an online task synthesis pipeline that creates tasks calibrated to the frontier of learnability for the current checkpoint. This report provides evidence that competitive small coding agents can be trained with synthetic tasks alone, without traditional distillation from larger models, and that generating tasks at the learnability frontier of the current agent is important. We report details on the training methodology, evaluations across diverse environments, and in-depth analyses, serving as a foundation for our ongoing exploration of lightweight yet capable coding agents that can run on minimal hardware.

While GUI agents have shown impressive capabilities in common computer-use tasks such as OSWorld, current benchmarks mainly focus on isolated and single-application tasks. This overlooks a critical real-world requirement of coordinating across multiple applications to accomplish complex profession-specific workflows. To bridge this gap, we present a computer-use benchmark in cross-application workflows, named WindowsWorld, designed to systematically assess GUI Agents on complex multi-step tasks that mirror real-world professional activities. Our methodology uses a multi-agent framework steered by 16 occupations to generate four difficulty-level tasks with intermediate inspection, which are then refined by human review and executed in a simulated environment. The resulting benchmark contains 181 tasks with an average of 5.0 sub-goals across 17 common desktop applications, of which 78% are inherently multi-application. Experimental results of leading large models and agents show that: 1) All computer-use agents perform poorly on multi-application tasks (< 21% success rate), far below the performance of simple single-app tasks; 2) They largely fail at tasks requiring conditional judgment and reasoning across geq 3 applications, stalling at early sub-goals; 3) Low execution efficiency, where tasks often fail despite far exceeding human step limits. Code, benchmark data, and evaluation resources are available at github.com/HITsz-TMG/WindowsWorld.

Computer-use agents hold the promise of assisting in a wide range of digital economic activities. However, current research has largely focused on short-horizon tasks over a limited set of software with limited economic value, such as basic e-commerce and OS-configuration tasks. A key reason is that creating environments for complex software requires significant time and human effort, and therefore does not scale. To address this, we introduce Gym-Anything, a framework for converting any software into an interactive computer-use environment. We frame environment creation itself as a multi-agent task: a coding agent writes setup scripts, downloads real-world data, and configures the software, while producing evidence of correct setup. An independent audit agent then verifies evidence for the environment setup against a quality checklist. Using a taxonomy of economically valuable occupations grounded in U.S. GDP data, we apply this pipeline to 200 software applications with broad occupational coverage. The result is CUA-World, a collection of over 10K long-horizon tasks spanning domains from medical science and astronomy to engineering and enterprise systems, each configured with realistic data along with train and test splits. CUA-World also includes CUA-World-Long, a challenging long-horizon benchmark with tasks often requiring over 500 steps, far exceeding existing benchmarks. Distilling successful trajectories from the training split into a 2B vision-language model outperforms models 2times its size. We also apply the same auditing principle at test time: a separate VLM reviews completed trajectories and provides feedback on what remains, improving Gemini-3-Flash on CUA-World-Long from 11.5% to 14.0%. We release all code, infrastructure, and benchmark data to facilitate future research in realistic computer-use agents.

Computer-using agents drive real software through the screen -- clicking and typing -- but they solve every task from scratch: asked to repeat a task, an agent re-reads the screen, re-reasons every tap, and pays the full cost again. We present PreAct, which lets such an agent get faster on tasks it has done before. The first time it succeeds, PreAct compiles the run into a small state-machine program-states that check the screen, transitions that act-and on later runs replays it directly instead of invoking the agent 8.5-13x faster, with no per-step language-model calls. Replay is not blind: at each step PreAct checks that the screen matches what the program expects before acting, and hands control back to the agent the moment something is off. PreAct applies the same discipline when deciding what to keep: a freshly compiled program enters the store only if, re-run from a clean state, an independent evaluator confirms it solved the task-catching programs that replay to their last step yet leave the task undone. Across a mobile, a desktop, and a web benchmark, this store-time check separates repeated runs that improve from ones that degrade as faulty programs accumulate, worth 1.75-2.6 tasks per benchmark, the same direction on all three; a fallback that explores afresh when no program fits brings PreAct level with a strong record-and-replay baseline. We also report what did not matter: prompt wording, runtime guardrails, and whether a language model or a plain embedding retriever selects which program to reuse.

Executable evaluation -- checking the consequences of an agent's actions with a program rather than grading its prose -- has become a prominent way to assess tool-using AI agents in software settings. Electric power engineering has not yet had an analogous benchmark: language-model use is still dominated by retrieval and text question answering, while agents acting on power-system artifacts remain mostly academic prototypes. We introduce the Power Systems Agent Benchmark, an executable benchmark for power-engineering agents. An agent receives a structured task and returns a structured solution; a deterministic evaluator recomputes the engineering quantities, checks operational constraints, and returns a feasibility flag, a normalized score, and explicit violations. The benchmark contains 41 task families across eight areas of power engineering, from power flow and protection to stability, microgrids, reliability, power quality, and forecasting. Each task is grounded in a citable source, standard, or documented engineering formulation. To resist contamination, held-out cases are synthesized on demand by per-family generators from private seeds: the construction is inspectable, but the instances remain private. In a reference evaluation with three command-line agents, the strongest score near the compact tier's ceiling, a smaller open model trails, and public and held-out performance are broadly consistent; a separate public-split grid with OpenCode and Aider probes harness effects. The reference evaluation doubles as quality control: unanimous failures flag candidate task or evaluator defects, and it exposed a latent evaluator bug missed by self-consistency checks. The evaluators are compact deterministic surrogates, but the task contract allows their internals to be upgraded to simulator-backed checks without changing how tasks are posed or solved.

Recent work in NLP has shown promising results in training models on large amounts of tasks to achieve better generalization. However, it is not well-understood how tasks are related, and how helpful training tasks can be chosen for a new task. In this work, we investigate whether knowing task relationships via pairwise task transfer improves choosing one or more source tasks that help to learn a new target task. We provide TaskWeb, a large-scale benchmark of pairwise task transfers for 22 NLP tasks using three different model types, sizes, and adaptation methods, spanning about 25,000 experiments. Then, we design a new method TaskShop based on our analysis of TaskWeb. TaskShop uses TaskWeb to estimate the benefit of using a source task for learning a new target task, and to choose a subset of helpful training tasks for multi-task training. Our method improves overall rankings and top-k precision of source tasks by 10% and 38%, respectively. We also use TaskShop to build much smaller multi-task training sets that improve zero-shot performances across 11 different target tasks by at least 4.3%.

Autonomous agents that execute human tasks by controlling computers can enhance human productivity and application accessibility. Yet, progress in this field will be driven by realistic and reproducible benchmarks. We present AndroidWorld, a fully functioning Android environment that provides reward signals for 116 programmatic task workflows across 20 real world Android applications. Unlike existing interactive environments, which provide a static test set, AndroidWorld dynamically constructs tasks that are parameterized and expressed in natural language in unlimited ways, thus enabling testing on a much larger and realistic suite of tasks. Reward signals are derived from the computer's system state, making them durable across task variations and extensible across different apps. To demonstrate AndroidWorld's benefits and mode of operation, we introduce a new computer control agent, M3A. M3A can complete 30.6% of the AndroidWorld's tasks, leaving ample room for future work. Furthermore, we adapt a popular desktop web agent to work on Android, which we find to be less effective on mobile, suggesting future research is needed to achieve universal, cross-domain agents. Finally, we conduct a robustness analysis by testing M3A against a range of task variations on a representative subset of tasks, demonstrating that variations in task parameters can significantly alter the complexity of a task and therefore an agent's performance, highlighting the importance of testing agents under diverse conditions. AndroidWorld and the experiments in this paper are available at https://github.com/google-research/android_world.

We introduce a principled probabilistic framework for reward-guided decoding in large language models, addressing the limitations of standard decoding methods that optimize token-level likelihood rather than sequence-level quality. Our method defines a reward-augmented target distribution over complete sequences by combining model transition probabilities with prefix-dependent reward potentials. Importantly, the approach is training-free: it leaves model weights unchanged and instead modifies the inference distribution via reward potentials, with all gains arising purely from inference-time sampling. To sample from this distribution, we develop Sequential Monte Carlo algorithms, including a computationally efficient prefix-only variant and a lookahead variant whose intermediate targets match the exact marginals of the full sequence distribution. The framework also integrates resample-move updates with Metropolis-Hastings rejuvenation and supports block-wise generation, subsuming common decoding strategies such as temperature sampling and power-tempered objectives. Empirical results across three 7B models show significant gains. On code generation (HumanEval), our method improves base performance by up to 54.9% and surpasses the strongest sampling baselines by 9.1%-15.3%. On mathematical reasoning (MATH500), it achieves gains of up to 8.8%. Notably, it reaches 87.8% on HumanEval and 78.4% on MATH500 with Qwen2.5-7B, consistently outperforming the reinforcement learning method GRPO.

LLM agents have demonstrated remarkable capabilities in software development, but their performance is hampered by long interaction contexts, which incur high API costs and latency. While various context compression approaches such as LongLLMLingua have emerged to tackle this challenge, they typically rely on fixed metrics such as PPL, ignoring the task-specific nature of code understanding. As a result, they frequently disrupt syntactic and logical structure and fail to retain critical implementation details. In this paper, we propose SWE-Pruner, a self-adaptive context pruning framework tailored for coding agents. Drawing inspiration from how human programmers "selectively skim" source code during development and debugging, SWE-Pruner performs task-aware adaptive pruning for long contexts. Given the current task, the agent formulates an explicit goal (e.g., "focus on error handling") as a hint to guide the pruning targets. A lightweight neural skimmer (0.6B parameters) is trained to dynamically select relevant lines from the surrounding context given the goal. Evaluations across four benchmarks and multiple models validate SWE-Pruner's effectiveness in various scenarios, achieving 23-54% token reduction on agent tasks like SWE-Bench Verified and up to 14.84x compression on single-turn tasks like LongCodeQA with minimal performance impact.

Beyond scratch coding, exploiting large-scale code repositories (e.g., GitHub) for practical tasks is vital in real-world software development, yet current benchmarks rarely evaluate code agents in such authentic, workflow-driven scenarios. To bridge this gap, we introduce GitTaskBench, a benchmark designed to systematically assess this capability via 54 realistic tasks across 7 modalities and 7 domains. Each task pairs a relevant repository with an automated, human-curated evaluation harness specifying practical success criteria. Beyond measuring execution and task success, we also propose the alpha-value metric to quantify the economic benefit of agent performance, which integrates task success rates, token cost, and average developer salaries. Experiments across three state-of-the-art agent frameworks with multiple advanced LLMs show that leveraging code repositories for complex task solving remains challenging: even the best-performing system, OpenHands+Claude 3.7, solves only 48.15% of tasks. Error analysis attributes over half of failures to seemingly mundane yet critical steps like environment setup and dependency resolution, highlighting the need for more robust workflow management and increased timeout preparedness. By releasing GitTaskBench, we aim to drive progress and attention toward repository-aware code reasoning, execution, and deployment -- moving agents closer to solving complex, end-to-end real-world tasks. The benchmark and code are open-sourced at https://github.com/QuantaAlpha/GitTaskBench.

One long-term goal of machine learning research is to produce methods that are applicable to reasoning and natural language, in particular building an intelligent dialogue agent. To measure progress towards that goal, we argue for the usefulness of a set of proxy tasks that evaluate reading comprehension via question answering. Our tasks measure understanding in several ways: whether a system is able to answer questions via chaining facts, simple induction, deduction and many more. The tasks are designed to be prerequisites for any system that aims to be capable of conversing with a human. We believe many existing learning systems can currently not solve them, and hence our aim is to classify these tasks into skill sets, so that researchers can identify (and then rectify) the failings of their systems. We also extend and improve the recently introduced Memory Networks model, and show it is able to solve some, but not all, of the tasks.

While long-horizon mobile GUI agents typically rely on thought-action-observation loops, they struggle to separate persistent task states from transient screen observations. As execution histories grow, this entanglement imposes a severe context burden, causing agents to forget initial requirements, hallucinate progress, or repeatedly interact with stale interfaces. To address this, we introduce Task-State Representation (TSR), a training-free framework that explicitly decouples task state from sensory input. Acting as a lightweight external wrapper, TSR maintains three structured components: a global instruction summary, a dynamic progress tracker for subgoals, and a transition-aware action verifier. By continuously updating through pre- and post-action visual comparisons, TSR effectively guides the agent's reasoning without requiring architectural modifications. Experiments across four mobile GUI benchmarks validate TSR's effectiveness, yielding up to a 12 absolute point increase in success rate on complex cross-application and memory-intensive tasks.

Existing computer-use agent benchmarks do not fully evaluate agents acting as assistants. A useful assistant retrieves information across complex, multi-step workflows, synthesizes it into artifacts (documents, presentations, spreadsheets), and navigates program interfaces to produce a coherent final product. Such workflows demand reasoning and synthesis, decomposition of complex tasks, as well as visual and spatial understanding. To study agents on workflows like these, we introduce KNOWS, a benchmark of open-ended, complex, browser-based tasks that jointly evaluate these capabilities, with each task culminating in a produced artifact. To write tasks, we develop a task design rubric and a protocol for ensuring that tasks meet the requirements. Each task is paired with an evaluator, a program that combines deterministic checks with LLM judgments to balance the richness, reliability, and automation tradeoff inherent to agent evaluation. We evaluate and analyze frontier computer-use agents and browser-based harnesses. They achieve moderate scores on partial-success metrics, but the best performer fully succeeds in fewer than 3% of our complex, long-horizon tasks. Failures on visual steps render the resulting artifacts unusable, even when agents complete more than 50% of other evaluation steps. Our results expose limitations of current agents acting as end-to-end assistants, and call for progress on tool use, visual understanding, and long-horizon reasoning.

Code models are increasingly trained with execution feedback, but most training signals still stop at correctness. This leaves an important gap for systems code: two programs can pass the same tests while differing greatly in runtime. We study how to train code agents to prefer faster correct implementations, rather than treating efficiency only as an evaluation metric. The key difficulty is that runtime is a fragile reward. It is meaningful only after a program is correct, varies across tasks, and gives little guidance when most sampled programs fail to compile or run. We propose RLPF, reinforcement learning from performance feedback, which turns execution outcomes into a staged reward. Failed programs are ordered by execution progress, while correct programs are ranked by their relative improvement from the baseline toward the expert reference. This gives useful feedback before correctness and performance-sensitive feedback after correctness. Fine-tuning Qwen3-32B with RLPF on PerfCodeBench raises correct-and-runnable solutions from 11.1% to 54.6% and improves relative efficiency from 8.1% to 38.6%. The trained model becomes competitive with stronger open-weight systems, and its optimization behavior transfers modestly to EffiBench-X. Additional studies show that model-generated references provide useful but weaker supervision, and that the full composite reward is more reliable than correctness-only or runtime-only baselines. These results suggest that code agents can be trained not only to pass tests, but also to optimize the programs they write.

Human feedback plays a pivotal role in aligning large language models (LLMs) with human preferences. However, such feedback is often noisy or inconsistent, which can degrade the quality of reward models and hinder alignment. While various automated data cleaning methods have been proposed to mitigate this issue, a systematic evaluation of their effectiveness and generalizability remains lacking. To bridge this gap, we introduce the first comprehensive benchmark for evaluating 13 preference data cleaning methods in the context of LLM alignment. PrefCleanBench offers a standardized protocol to assess cleaning strategies in terms of alignment performance and generalizability across diverse datasets, model architectures, and optimization algorithms. By unifying disparate methods and rigorously comparing them, we uncover key factors that determine the success of data cleaning in alignment tasks. This benchmark lays the groundwork for principled and reproducible approaches to improving LLM alignment through better data quality-highlighting the crucial but underexplored role of data preprocessing in responsible AI development. We release modular implementations of all methods to catalyze further research: https://github.com/deeplearning-wisc/PrefCleanBench.

The overarching objective of this paper is two-fold. First, to explore model-based approaches to characterize the primary cause of the noise. in the RE dataset TACRED Second, to identify the potentially noisy instances. Towards the first objective, we analyze predictions and performance of state-of-the-art (SOTA) models to identify the root cause of noise in the dataset. Our analysis of TACRED shows that the majority of the noise in the dataset originates from the instances labeled as no-relation which are negative examples. For the second objective, we explore two nearest-neighbor-based strategies to automatically identify potentially noisy examples for elimination and reannotation. Our first strategy, referred to as Intrinsic Strategy (IS), is based on the assumption that positive examples are clean. Thus, we have used false-negative predictions to identify noisy negative examples. Whereas, our second approach, referred to as Extrinsic Strategy, is based on using a clean subset of the dataset to identify potentially noisy negative examples. Finally, we retrained the SOTA models on the eliminated and reannotated dataset. Our empirical results based on two SOTA models trained on TACRED-E following the IS show an average 4% F1-score improvement, whereas reannotation (TACRED-R) does not improve the original results. However, following ES, SOTA models show the average F1-score improvement of 3.8% and 4.4% when trained on respective eliminated (TACRED-EN) and reannotated (TACRED-RN) datasets respectively. We further extended the ES for cleaning positive examples as well, which resulted in an average performance improvement of 5.8% and 5.6% for the eliminated (TACRED-ENP) and reannotated (TACRED-RNP) datasets respectively.

With the rapid development of large language models (LLMs), the quality of training data has become crucial. Among the various types of training data, mathematical data plays a key role in enabling LLMs to acquire strong reasoning abilities. While high-quality open-source data is important, it is often insufficient for pre-training, necessitating the addition of synthetic math problems. However, synthetic math questions and answers can introduce inaccuracies, which may degrade both the training data and web data. Therefore, an effective method for cleaning synthetic math data is essential. In this paper, we propose the MathClean benchmark to evaluate the effectiveness of math data cleaning models. The MathClean benchmark consists of 2,000 correct questions and 2,000 erroneous questions with additional 2,000 correct and erroneous answers sourced from augmented data based on GSM8K and MATH. Moreover, we also annotate error types for each question or answer, since it can assess whether models can correctly identify the error categories for future improvements. Finally, we present comprehensive evaluations using state-of-the-art (SOTA) models. Our results demonstrate that even strong models like GPT-o1 and DeepSeek-R1 perform poorly on this benchmark, highlighting the utility of MathClean. Our code and data is available at https://github.com/YuYingLi0/MathClean.

We provide the first formal definition of reward hacking, a phenomenon where optimizing an imperfect proxy reward function leads to poor performance according to the true reward function. We say that a proxy is unhackable if increasing the expected proxy return can never decrease the expected true return. Intuitively, it might be possible to create an unhackable proxy by leaving some terms out of the reward function (making it "narrower") or overlooking fine-grained distinctions between roughly equivalent outcomes, but we show this is usually not the case. A key insight is that the linearity of reward (in state-action visit counts) makes unhackability a very strong condition. In particular, for the set of all stochastic policies, two reward functions can only be unhackable if one of them is constant. We thus turn our attention to deterministic policies and finite sets of stochastic policies, where non-trivial unhackable pairs always exist, and establish necessary and sufficient conditions for the existence of simplifications, an important special case of unhackability. Our results reveal a tension between using reward functions to specify narrow tasks and aligning AI systems with human values.

Agent benchmarks often evaluate only final answers even when agents run on stateful runtimes. We argue this under-specifies what is being evaluated: the proper unit is a declared model-plus-runtime configuration whose failures can occur in evidence acquisition, runtime routing, safety boundaries, or repeated execution. We present ClawProBench, a trace-aware benchmark for runtime-native agent evaluation instantiated on OpenClaw, a live agent runtime with workspace tools and native surfaces for browsing, memory, messaging, scheduling, skills, and subagents. ClawProBench defines two tracks: a 102-scenario full profile with live workspace and native-runtime routing tasks, and a frozen 68-scenario holdout with closed-world JSON output contracts for robust ranking. Trials are scored from execution traces via a safety-gated formula combining correctness, process quality, and efficiency, preserving failure evidence for audit. Our anonymous artifact includes benchmark definitions, scoring code, manifests and sanitized traces. We evaluate 68 configurations on the full profile and 37 on holdout. The top safety-gated average trace score is 0.7671. Native-runtime tasks underperform workspace-live tasks (0.5238 vs. 0.6415). On holdout, pass@k-any outperforms strict three-trial pass (0.6638 vs. 0.2890), while full-profile and holdout rankings show weak alignment (Spearman 0.1300). Rankings based purely on correctness differ substantially from process-aware, safety-gated and strict-pass views. Final-answer leaderboards may hide native-surface weaknesses, one-off successes and trace-local agent failure modes.

Household tidying is an important application area, yet current benchmarks neither model user preferences nor support mobility, and they generalize poorly, making it hard to comprehensively assess integrated language-to-action capabilities. To address this, we propose RoboTidy, a unified benchmark for language-guided household tidying that supports Vision-Language-Action (VLA) and Vision-Language-Navigation (VLN) training and evaluation. RoboTidy provides 500 photorealistic 3D Gaussian Splatting (3DGS) household scenes (covering 500 objects and containers) with collisions, formulates tidying as an "Action (Object, Container)" list, and supplies 6.4k high-quality manipulation demonstration trajectories and 1.5k naviagtion trajectories to support both few-shot and large-scale training. We also deploy RoboTidy in the real world for object tidying, establishing an end-to-end benchmark for household tidying. RoboTidy offers a scalable platform and bridges a key gap in embodied AI by enabling holistic and realistic evaluation of language-guided robots.

Computer-using agents (CUAs) enable task completion through natural interaction with operating systems and software interfaces. While script-based verifiers are widely adopted for evaluation, they suffer from limited scalability and inability to provide step-wise assessment. Reward models offer promising alternatives, but their effectiveness on CUA evaluation remains largely underexplored. To address this gap, we present CUARewardBench, comprising four key contributions: (1) First-ever Comprehensive CUA Reward Benchmark: We introduce the first benchmark for evaluating both outcome reward models (ORM) and process reward models (PRM) on CUA tasks, enabling systematic assessment across trajectory-level and step-level evaluation. (2) Diverse, Practical and Reliable Dataset: CUARewardBench encompasses trajectories from 10 software categories and 7 agent architectures with varying performance levels (25.9%-50.8% success rates). All trajectories are expertly annotated through carefully designed protocols, with rigorous quality control to ensure reliability and practical applicability. (3) Comprehensive Analysis and Insights: Through extensive experiments across 7 vision-language models and 3 prompt templates, we reveal critical limitations of current CUA RMs, including insufficient visual reasoning capabilities, knowledge deficiencies, and the superiority of general VLMs over specialized CUA models for reward evaluation. (4) Unanimous Prompt Ensemble (UPE): Based on the insights from our comprehensive analysis, we propose UPE, a novel ensemble method that significantly enhances reward model reliability through strict unanimous voting and strategic prompt-template configurations. UPE achieves 89.8% precision and 93.3% NPV for ORM, and 81.7% precision and 85.1% NPV for PRM, substantially outperforming single VLMs and traditional ensemble approaches.

Desktop interaction streams provide a continuous, privacy-sensitive record of interleaved user tasks. Transforming these streams into task-organized personal logs on-device faces two main challenges: exhaustive Vision-Language Model (VLM) processing strains local resources, and global stream processing causes cross-task context pollution. We present FOCAL (Filtered On-device Continuous Activity Logging), a privacy-first multi-agent system utilizing a unified filter-plan-log architecture. It cascades a lightweight Filter Agent for noise suppression, a text-only Brain Agent for task attribution, a Record Agent for selective visual reasoning, and a task-isolated Memory Agent for context-coherent summarization. Experiments on DesktopBench (comprising 2,572 screenshots across 420 complex sessions) show FOCAL reduces total token consumption by 60.4% and VLM call count by 72.3% versus a baseline, while boosting Key Information Recall (KIR) from 0.38 to 0.61. Crucially, under A{to}B{to}A task interruptions, FOCAL maintains Task Acc 0.81 and KIR 0.80, whereas the baseline collapses to Task Acc 0.03. FOCAL pioneers the efficient, on-device summarization of instruction-free desktop streams into multi-perspective personal logs.

One of the central tasks in software maintenance is being able to understand and develop code changes. Thus, given a natural language description of the desired new operation of a function, an agent (human or AI) might be asked to generate the set of edits to that function to implement the desired new operation; likewise, given a set of edits to a function, an agent might be asked to generate a changed description, of that function's new workings. Thus, there is an incentive to train a neural model for change-related tasks. Motivated by this, we offer a new, "natural", large dataset of coupled changes to code and documentation mined from actual high-quality GitHub projects, where each sample represents a single commit where the code and the associated docstring were changed together. We present the methodology for gathering the dataset, and some sample, challenging (but realistic) tasks where our dataset provides opportunities for both learning and evaluation. We find that current models (specifically Llama-3.1 405B, Mixtral 8times22B) do find these maintenance-related tasks challenging.

Coding-agent benchmarks are built for broad and comprehensive comparisons, not frequent development decisions. Individual runs vary, full suites are expensive, and the benchmark harness may differ from the harness used in practice. In a resampling analysis of DeepSWE's published trials, only 19.5% of tasks (22 of 113) had a fifth-percentile Pearson correlation of at least 0.50 with full-benchmark performance. The paper presents DeltaSelect, an open-source method that identifies tasks whose one-run results consistently track full-benchmark performance using Pearson correlation, maps fractional verifier results to a common score using linear regression, and selects a fixed task set within a dollar budget. DeltaSelect is intended for repeated baseline-versus-candidate comparisons during development, not model rankings. In a gpt-5.6-luna low-reasoning case study, DeltaSelect was used to revise custom skills and instructions. Across 13 evaluations, the recorded cost was USD 27.86 at rates published August 16, 2026. The adopted version cost 58.1% less than the initial version (USD 1.75 versus USD 4.18; p=0.008), while the calibrated score was higher (42.36% versus 36.46%; published-analog variance p=0.326).

We introduce GDPval, a benchmark evaluating AI model capabilities on real-world economically valuable tasks. GDPval covers the majority of U.S. Bureau of Labor Statistics Work Activities for 44 occupations across the top 9 sectors contributing to U.S. GDP (Gross Domestic Product). Tasks are constructed from the representative work of industry professionals with an average of 14 years of experience. We find that frontier model performance on GDPval is improving roughly linearly over time, and that the current best frontier models are approaching industry experts in deliverable quality. We analyze the potential for frontier models, when paired with human oversight, to perform GDPval tasks cheaper and faster than unaided experts. We also demonstrate that increased reasoning effort, increased task context, and increased scaffolding improves model performance on GDPval. Finally, we open-source a gold subset of 220 tasks and provide a public automated grading service at evals.openai.com to facilitate future research in understanding real-world model capabilities.