Benchmark Environment

Small leaderboard gaps are often interpreted as evidence that one language model is better than another, but their sign may depend on which benchmark items are included. We test this using item-level responses from five benchmarks and a family-label-free spectral approximation to multidimensional item-response theory (MIRT). In owner-disjoint folds, one owner half identifies items with low residual differential item functioning across model families (low-DIF); the resulting frozen, source- and easiness-balanced weights score models in the other half, while equally short matched-random subtests control for generic subtest variation. Full-benchmark and low-DIF rankings remain strongly correlated ($τ_b=.900$--$.948$). Yet in four of five benchmarks, 30.9--47.1% of cross-family pairs initially within one percentage point reverse order, exceeding their matched-random medians by 16.9--28.6 percentage points (all $p=.001$). The fifth benchmark shows no reliable excess ($-0.9$ points, $p=.689$). The pattern survives all pre-specified population perturbations, and residual item--family signatures replicate across owner halves; however, no family shows a consistent advantage across benchmarks. Thus, globally stable rankings can still leave individual near-tie orderings sensitive to benchmark composition, and sub-one-point leaderboard gaps should be accompanied by evidence that the implied ordering is composition-robust.

Efficient evaluation changes the protocol used to support claims about model behavior, yet it is rarely tested whether those claims remain stable after the evaluation itself is made cheaper. We stress-test conclusion robustness in responsible-AI benchmarking by evaluating three dense and mixture-of-experts models on BBQ and BBQ-V under seven conditions spanning batching, quantization, benchmark reduction, and their combinations. Rather than treating preserved aggregate accuracy as sufficient, we compare accuracy, bias severity and prevalence, reasoning quality, subgroup behavior, subset-membership stability, runtime, and measured GPU energy against a full-benchmark BF16 baseline. Larger batching keeps accuracy within 0.35 percentage points of baseline and produces comparatively small subgroup changes, while reducing energy in five of six model--dataset settings. INT8 largely preserves quality but uses 1.79--4.26$\times$ baseline energy. INT4 causes larger, model- and context-dependent changes. Reduced benchmarks provide the most consistent savings, but very small subsets are substantially more sensitive to which items are retained. Efficient evaluation should therefore be treated as a measurement intervention whose validity must be checked across the conclusions the benchmark is intended to support. Our project website is https://vectorinstitute.github.io/sustainable-rai-evaluation/ and the code is available at https://github.com/VectorInstitute/sustainable-rai-evaluation.

Automated maritime surveillance from satellite and aerial imagery requires large, precisely annotated datasets, which remain scarce for the instance-segmentation task, particularly for small vessels in cluttered port environments. We present MariSat, a new benchmark dataset of 1260 aerial and satellite images covering diverse port and coastal scenes, annotated at the pixel level for eight maritime object classes (sailboat, yacht, jet-ski, fishing boat, cruise ship, military vessel, tugboat and cargo ship). The dataset was produced through a semi-automatic annotation pipeline combining the textpromptable segmentation model SAM 3 with a cascade of geometric and colorimetric post-processing filters, followed by a manual correction and quality-control pass performed with the CVAT annotation platform. We describe the image-collection methodology, the annotation and correction process, and the resulting data organization. We also report class-wise statistics for the training, validation, and test splits. MariSat has already been used to fine-tune and benchmark segmentation and detection models (SAM 3 and YOLO11) for real-time maritime monitoring. We report detailed quantitative and per-class results for both tasks. The MariSat dataset is publicly available on GitHub : https://github.com/amirabbes/P2M-Maritime-Segmentation

Multi-agent systems (MAS) built on Large Language Models (LLMs) are proliferating rapidly, but their heterogeneous execution traces provide no common basis for evaluation across methods. Outcome-only benchmarks discard collaborations, whereas LLM-as-Judge evaluation requires additional, model-dependent inference and can vary with the LLM and rubric. We introduce a generalizable evaluation framework that maps native MAS traces into a shared space of unified collaboration graphs, enabling different methods to be evaluated under the same representation, reference set, and metric panel. Candidate graphs are compared with a query-specific reference forest. Each forest is a benchmark-provided collection of verified-success graphs: it records diverse ways in which representative MAS methods can complete the task, rather than prescribing a unique optimal process. Instantiating the framework as ForestBench, we filter $844$ collaboration-necessary queries from seven public datasets, precompute ten successful target-conditioned reference graphs per query, and evaluate six representative MAS frameworks. Controlled backbone, reference-construction, and perturbation studies test the stability and scope of evaluation. Once the benchmark forests are built, ForestBench scores a trace in milliseconds without further LLM inference, providing a reusable structural basis for comparing diverse MAS collaboration traces.

Revenue forecasts guide acquisition budgets, demand planning, and customer-based valuations, yet an aggregate forecast does not show whether change reflects acquisition, repeat purchasing, spending per order, or offsetting movements. Using weekly transaction panels for 966 companies in 25 industries, the authors develop the Customer-Based Multi-task Transformer (CBMT), which learns shared structure, retains separate primitive forecasts, and aligns their combination with downstream revenue. CBMT's mean total-sales error is 30% below the strongest representative established customer-base benchmark. It is also 2.65% below a Transformer that forecasts total sales directly, although the paired difference is not statistically significant (p=.222), and it beats separately estimated single-task forecasts for 74.3% of firms. CBMT's source MAE is lower in 23 of 24 benchmark-by-outcome comparisons, with the remaining difference not statistically distinguishable from zero. Firms whose primitives co-move more strongly are more likely to benefit from joint forecasting; selected-family scenario-3 comparisons are consistent with gains from shared representation and revenue alignment but remain diagnostic rather than causal. Accuracy deteriorates for all models when customer-base dynamics are highly volatile, and CBMT's advantage narrows there. Calibration-period routing rules do not improve average accuracy over always deploying CBMT. The results show how coordinated customer-base forecasts support revenue planning and when they warrant greater caution.

LLM-agent evaluations often produce task outcomes long before the full benchmark run is complete. A partial score is tempting to report, but it does not show whether the observed tasks support the same conclusion as the completed evaluation. Early tasks can omit important parts of a benchmark, running cheaper tasks first can distort the observed sample, and a rule that decides only easy pairs can appear accurate while leaving many comparisons unresolved. We introduce ParEvalLayer, a decision layer that reads paired outcomes for two agent systems and a comparison policy chosen in advance. For each partial run, it records whether the tested agent system is better by the required amount, is not better by that amount, needs more evidence, or should abstain. We evaluate ParEvalLayer by replaying completed public benchmark data as if each evaluation had stopped earlier. At each point, ParEvalLayer applies the policy using only the outcomes observed so far; if it reaches one of the two comparison judgments, we check whether that judgment matches the completed data for the same system pair. With the main comparison rule, three of the public benchmarks reach the same decision as the completed evaluation after observing only 15% to 25% of task outcomes. Other benchmarks require more task outcomes. This variation shows why a partial score alone is not enough: reports should also state the decision rule and how many comparisons remain without a decision.

Unified humanoid policies handle agile whole-body motion, yet stumble on a simple demand: staying balanced on one leg. On our single-leg-balance benchmark, eight released state-of-the-art general policies hold a clean single-leg stance on 0 of 90 test motions; they stay up only by stepping or hopping, recovering from imbalance rather than preventing it. Prevention needs the capture point (xCoM), the center of mass (CoM) extrapolated by its velocity, which has never driven a hardware policy because it requires a base linear velocity no on-board sensor provides; expressed relative to the support foot, that velocity cancels exactly, leaving an observation reconstructible from encoders and IMU alone. We put this first deployable dynamic-CoM observation directly into the actor that runs on hardware, and pair it with a reward library translated term by term from human postural control, under one principle: prevention over repair. Trained by asymmetric FastSAC with a privileged critic and no distillation, the resulting policy, FDDC (First Deployable Dynamic-CoM), holds clean single-leg balance on 86 of 90 held-out motions across nine stratified pose classes and transfers to a real Unitree G1; in ablation, the dynamic-CoM observation is the single largest driver: removing it alone costs 40 points of clean single-leg balance. We release the full stack with the first method-agnostic, reproducible sim2sim benchmark for humanoid single-leg balance, scoring each policy in a simulator distinct from its training one, a step toward turning balance from a per-task trick into a capability the field can measure.

Unified audio models capable of audio understanding, audio generation and, increasingly, audio editing are proliferating rapidly. Yet a basic question about them remains unanswered: do the two heads of a unified model agree about the same audio? Current practice evaluates each capability in isolation on specialized benchmarks, and never asks whether a model can make sense of its own generations. We present TORUS, the first self-coherence test for audio-native unified models. TORUS comprises 48 three-stage self-coherence tests carrying 432 six-option questions spanning speech, sound and music across five task families. We holistically evaluate five open unified models alongside a Cascaded Baseline that combines state-of-the-art specialized generation, editing and understanding models. The best unified model answers 50.5% of questions against the Cascaded Baseline's 63.2% and a 16.7% chance floor. Models struggle on audio editing. Among the evaluated audio models (specialized and unified), we observe limited self-coherence, and thus position self-coherence as an essential test for future audio systems.

As large language models (LLMs) are deployed as agents in high-stakes settings, such as medical and legal systems, understanding their deceptive capabilities is fundamental to safety. Controlled social deduction games provide a reproducible proxy for isolating and evaluating these complex adversarial behaviors. We present the open-source benchmark framework ParliamentBench based on the game Secret Hitler to evaluate LLMs in scenarios that require deception, persuasion, and reasoning under information asymmetry. We evaluate 16 LLMs across 1,600 simulated matches playing each other, playing against humans, and compare them against a large set of online games. We introduce three novel metrics that isolate social deduction, reasoning, and deceptive consistency. Our experiments reveal that frontier models achieve strong performance across cooperative and deceptive roles, with a strong top-four cluster (GPT-5.4, Kimi K2.5, Grok 4.1 Fast, and DeepSeek 3.1 Terminus), whereas the weakest models fall short of random (33%) and simple algorithmic (45%) baselines. Most LLMs struggle to maintain a consistent deceptive persona throughout an entire game, with deception retention dropping below 50%.

Offline root-cause-analysis (RCA) benchmarks commonly rank methods by a single pooled top-1 accuracy across multiple subsystems, and engineers often read the pooled winner as a recommendation for their own subsystem. We audit that reading on three public RCA benchmark families -- OpenRCA, RCAEval, and PetShop -- covering 11 subsystems and 778 matched scoring units. To keep pairwise comparisons on identical cases, the main analysis retains four methods or comparators with complete coverage: BARO, a CD-1min adapter, max-$|Z|$, and per-service alert-count. All six pairwise comparisons show subsystem-level effects of both signs, every random-effects 95% prediction interval crosses zero, and case-level interaction tests reject exchangeability in 5 of 6 pairs. Leave-one-system-out selection picks the lower-scoring method on up to 5 of 11 held-out subsystems, with regret reaching 24.8 pp on RCAEval / Sock-Shop. We release a 320-line audit module; given a matched RCA benchmark score table, it recomputes the same per-subsystem stability checks alongside pooled scores.

Static visual question answering (VQA) benchmarks age quickly: Once the items leak into training corpora, scores can reflect memorization rather than genuine visual ability, thus obscuring real progress. Rebuilding high-quality benchmarks such as VBench requires substantial human annotation, yet each static release can quickly become another leaked artifact. We propose ReKey, a live benchmark protocol that randomly regenerates the answer-bearing local detail, or visual key, in real images at evaluation time. Using human-validated edit slots, ReKey samples fresh instances with new answers, construction-grounded labels, and controlled visual-search difficulty. On VBench, the ReKey regenerated benchmark reveals a sharp score jump across eight frontier vision-language models (VLMs): The original items score 9.5--18.8 percentage points higher than the regenerated variants. By making the visual key renewable, ReKey keeps evaluation fresh as models and training data evolve.

Algorithmic recourse methods provide counterfactual explanations that inform individuals of the actions required to overturn an unfavorable model decision. Despite rapid methodological progress, principled comparison remains elusive; existing frameworks are often difficult to extend and lack both interoperability and systematic verification that integrated methods faithfully reproduce their originally reported results. We introduce \emph{RecourseBench}, a unified evaluation framework built around three commitments namely, modularity, reproducibility, and interactivity. The framework decomposes the pipeline into five fully decoupled layers -- Data, Preprocessing, Model, Recourse Method, and Evaluation -- governed by abstract interfaces and a dynamic registry. To address the reproducibility gap in prior benchmarks, we introduce a four-tier classification system in which every integrated method is validated by an automated test suite against its originally reported results. We further provide an interactive web interface for flexible, configuration-driven comparison across methods, datasets, and model architectures. Our framework currently integrates 28 state-of-the-art recourse methods and, to our knowledge, constitutes the first recourse benchmark to explicitly enforce method-level reproducibility through automated, quantitative testing.

In reactor physics, neutronics can be treated with different fidelity levels, according to the needs of the user. On one hand, the precise modeling of neutrons' behaviour in reactor physics is often expensive and time-consuming due to the high computational costs to numerically solve the Boltzmann transport equation. Conversely, by adopting suitable assumptions, such as the SP$_N$, diffusion theory, and point kinetics, it is possible to generate efficiently low-fidelity data. From the perspective of surrogate models, this computational limitation translates into a scarcity of high-fidelity data and a significant amount of low-fidelity data. Given this difference in fidelity levels, it would be interesting to develop a suitable procedure to map low-fidelity models towards higher fidelity models; for instance, one could obtain the solution to a multi-group diffusion equation starting from time-series data obtained from a point kinetics model. Indeed, this work investigates this possibility by leveraging multi-fidelity information with Shallow Recurrent Decoders, a novel machine learning architecture able to map time-series observations to the full state of the reactor. This technique has been designed to use local or global measurements as input and map their temporal trajectories to the high-dimensional state; by the same logic, in principle, this architecture can also be used when the input is formed by the solution of a lumped model. This work applies this idea to a benchmark reactor geometry, mapping the point kinetics model to the diffusion solution under various input conditions, with much less computational costs.

Generative models have demonstrated remarkable potential in time series analysis tasks, like synthesis, forecasting, imputation, etc. However, offering limited coverage for generative models, existing time series libraries are mainly engineered for discriminative models, with standardized workflows for specific tasks, such as optimizing Mean Squared Errors for time series forecasting. This rigid structure is fundamentally incompatible with the distinct and often complex paradigms of generative models (e.g., adversarial training, diffusion processes), which learn the underlying data distribution rather than a direct input-output mapping. To this end, we proposed GenTS, a comprehensive and extensible benchmark library designed for systematic assessment on generative time series models. GenTS features a unified data preprocessing pipeline, a collection of versatile models, and panoramic evaluation metrics. Its modular design also enables the researchers to flexibly customize beyond our built-in datasets and models. Based on GenTS, we conducted benchmarking experiments under diverse tasks, accordingly offering suggestions for model selection and identifying potential directions for future research. Our codes are open-source at https://github.com/WillWang1113/GenTS. The official tutorials and document are available at https://willwang1113.github.io/GenTS/.

Every AI benchmark operationalizes theoretical assumptions about the capability it claims to assess. When assumptions function as unexamined commitments, benchmarks stabilize the dominant paradigm by narrowing what counts as progress. Over time, narrow evaluation reorganizes capability concepts: architectures and definitions are selected for benchmark legibility until evaluation ceases to track an independent object and instead produces a version of the target defined by its own operational assumptions. The result is a trap: evaluation frameworks treat self-reinforcing assessments as valid, both creating and obscuring structural limits on what the current paradigm can accomplish. We introduce Epistematics, a methodology for deriving evaluation criteria directly from technical capability claims and auditing whether proposed benchmarks can discriminate the claimed capability from proxy behaviors. The contribution is meta-evaluative: an audit procedure, a failure mode taxonomy, and benchmark-design criteria for evaluating capability-evaluation coherence. We demonstrate the procedure through a worked audit of Dupoux et al. (2026), a proposal that revises the dominant paradigm's theoretical assumptions at the architectural level while reproducing them in its evaluation criteria, thereby entrenching the constraint it seeks to overcome in a form the evaluation cannot detect.

The primary way to establish and compare competencies in foundation and generative AI models has shifted from peer-reviewed literature to press releases and company blog posts, where model builders highlight results on selected benchmarks. These artifacts now largely define the state of the art for researchers and the public. Despite their prominence, which benchmarks model builders choose to highlight, and what they communicate through this selection, is underexamined. To investigate, we introduce and open-source Benchmarking-Cultures-25, a dataset of 231 benchmarks highlighted across 139 model releases in 2025 from 11 major AI builders, alongside an interactive tool to explore the data. Our analysis reveals a fragmented evaluation landscape with limited cross-model comparability: 63.2% of highlighted benchmarks are used by a single builder, and 38.5% appear in just one release. Few achieve widespread use (e.g., GPQA Diamond, LiveCodeBench, AIME 2025). Moreover, benchmarks are attributed different competencies by different builders, depending on their narrative. To disentangle these conflicting presentations, we develop a unified taxonomy mapping diverging terminology to a shared framework of measured signals based on what benchmark authors claim to measure. "General knowledge application" is the second most popular, yet vaguely defined, category. Qualitative analysis shows many such benchmarks deemphasize construct validity, instead framing results as indicators of progress toward AGI. Their authors claim to measure knowledge or reasoning broadly, yet mostly evaluate STEM subjects (especially math). We argue that highlighted benchmarks function less as standardized measurement tools and more as flexible narrative devices prioritizing market positioning over scientific evaluation. Data: https://hf.co/datasets/matybohacek/benchmarking-cultures-25; tool: https://bench-cultures.net.

Interactive narrative tasks require LLMs to sustain a coherent, evolving story while adapting to a user over multiple turns. However, suitable benchmarks for this setting are limited: existing evaluations often focus on static prompts, isolated story generations, or post-hoc ratings, and therefore miss whether models can jointly manage story generation, long-context state and pacing, character simulation, empathic personalization, and story-grounded artifacts. We introduce NARRA-Gym, an executable evaluation environment that turns a sparse emotional seed into a complete interactive story episode and logs the full model-in-the-loop trajectory, including story construction, memory updates, planning, pacing interventions, and optional artifact synthesis. We evaluate nine frontier LLMs using a controlled LLM-as-judge sweep over eight benchmark personas and a human evaluation in which participants rate customized model outputs. Our results show substantial variation across models, personas, and evaluation dimensions: models that produce fluent stories can still fail on robustness, user experience, or resistance-sensitive personalization. These findings suggest that interactive narrative offers a useful benchmark for evaluating long-horizon, user-adaptive LLM behavior beyond isolated story quality.

Stopping criteria automatically determine when to stop an evolutionary algorithm, so as not to waste function evaluations on a stagnant population. Although stopping criteria play an important role in real-world applications, they have attracted little attention in the evolutionary multi-objective optimization (EMO) community. In fact, new stopping criteria for EMO have been rarely developed in recent years. One reason for the stagnation in developing stopping criteria for EMO is a lack of effective benchmarking methodologies. To address this issue, this paper proposes (i) a performance measure of stopping criteria for EMO and (ii) a file-based benchmarking approach. This paper also proposes (iii) a data representation method that effectively stores population states in text files. (i) The proposed measure represents the performance of stopping criteria as a single scalar value, making comparison easy. (ii) The proposed file-based approach not only simplifies the benchmarking process but also facilitates reproducibility. (iii) The proposed data representation method addresses the issue of file size in (ii). We demonstrate the effectiveness of our three contributions (i)--(iii) by benchmarking five representative stopping criteria for EMO.

The rigorous evaluation of the novelty of a scientific paper is, even for human scientists, a challenging task. With the increasing interest in AI scientists and AI involvement in scientific idea generation and paper writing, it also becomes increasingly important that this task be automatable and reliable, lest both human attention and compute tokens be wasted on ideas that have already been explored. Due to the challenge of quantifying ground-truth novelty, however, existing novelty metrics for scientific papers generally validate their results against noisy, confounded signals such as citation counts or peer review scores. These proxies can conflate novelty with impact, quality, or reviewer preference, which in turn makes it harder to assess how well a given metric actually evaluates novelty. We therefore propose an axiomatic benchmark for scientific novelty metrics. We first define a set of axioms that a well-behaved novelty metric should satisfy, grounded in human scientific norms and practice, then evaluate existing metrics across ten tasks spanning three domains of AI research. Our results reveal that no existing metric satisfies all axioms consistently, and that metrics fail on systematically different axioms, reflecting their underlying architectures. Additionally, we show that combining metrics of complementary architectures leads to consistent improvements on the benchmark, with per-axiom weighting achieving 90.1% versus 71.5% for the best individual metric, suggesting that developing architecturally diverse metrics is a promising direction for future work. We release the benchmark code as supplementary material to encourage the development of more robust scientific literature novelty metrics.

Recent advances in large language models (LLMs) have enabled agentic systems to translate natural-language intent into executable scientific visualization (SciVis) tasks. Despite rapid progress, the community lacks a principled and reproducible benchmark for evaluating these emerging SciVis agents in realistic, multi-step analysis settings. We present SciVisAgentBench, a comprehensive and extensible benchmark for evaluating scientific data analysis and visualization agents. Our benchmark is grounded in a structured taxonomy spanning four dimensions: application domain, data type, complexity level, and visualization operation. It currently comprises 108 expert-crafted cases covering diverse SciVis scenarios. To enable reliable assessment, we introduce a multimodal outcome-centric evaluation pipeline that combines LLM-based judging with deterministic evaluators, including image-based metrics, code checkers, rule-based verifiers, and case-specific evaluators. We also conduct a validity study with 12 SciVis experts to examine the agreement between human and LLM judges. Using this framework, we evaluate representative SciVis agents and general-purpose coding agents to establish initial baselines and reveal capability gaps. SciVisAgentBench is designed as a living benchmark to support systematic comparison, diagnose failure modes, and drive progress in agentic SciVis. The benchmark is available at https://scivisagentbench.github.io/.

Preference-Conditioned Policy Learning (PCPL) in Multi-Objective Reinforcement Learning (MORL) approximates diverse Pareto-optimal solutions by conditioning a single policy on user-specified preferences, enabling run-time adaptation to arbitrary trade-offs without retraining. However, existing PCPL benchmarks are largely restricted to toy tasks and fixed environments, limiting their realism and scalability. To address this gap, we introduce GraphAllocBench, a flexible benchmark built on CityPlannerEnv, a novel graph-based resource allocation sandbox inspired by city management. GraphAllocBench provides a rich suite of problems with customizable objective functions, varying preference conditions, complex Pareto Fronts, and high-dimensional scalability. We further propose two supplementary metrics -- Proportion of Non-Dominated Solutions (PNDS) and Ordering Score (OS) -- that capture prediction reliability and preference consistency while complementing the widely used hypervolume metric. Through experiments with several state-of-the-art PCPL algorithms and our own MLP and graph-aware PCPL-PPO baseline, we show that GraphAllocBench exposes distinct failure modes that hypervolume alone does not capture but our supplementary metrics reveal, while motivating graph-based approaches such as Graph Neural Networks (GNNs) for scaling to complex, high-dimensional allocation tasks. By letting users freely vary objectives, preferences, and allocation rules, GraphAllocBench serves as a versatile and extensible testbed for advancing PCPL.

PCB routing is the task of connecting the nets of a board with copper traces under strict design rules, yet learning-based methods still lag behind rule-based routers. We introduce PCBWorld, an open-source engine-grounded PCB routing environment built on KiCad, an electronic design automation (EDA) engine. As a human engineer does, agents in PCBWorld interactively route a board through the engine's native operations, guided by its Design Rule Check (DRC) feedback. The environment supports both RL and tool-using LLM agents. Alongside the environment, PCBWorld-Bench provides three board datasets in the native .kicad_pcb format, two controllable synthetic generators and 679 real open-source boards. It scores any completed board with eight engine-checked evaluation metrics, regardless of the routing method. In our experiments, agents in PCBWorld consistently outperformed grid-action RL policies and open-loop LLM baselines, and an RL policy trained only on synthetic boards transferred zero-shot to real boards, approaching rule-based routers. PCBWorld and PCBWorld-Bench are available at https://github.com/LGAI-Research/PCBWorld.