LLM Agent Evaluation

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

115 papers in the last four weeks, up 140% on the four weeks before. 1.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 793

Jun 1, 2026cs.AI

BADGER: Bridging Agentic and Deterministic Evaluation for Generative Enterprise Reasoning

Enterprise AI systems that translate natural language into SQL queries and orchestrate multi-step agentic reasoning pipelines require evaluation approaches fundamentally different from academic benchmarks. Spider and BIRD established execution-accuracy protocols; G-Eval and RAGAS advanced LLM-based assessment; and recent work such as Spider 2.0, BEAVER, and BIRD-Interact has begun to address enterprise and agentic dimensions. No single framework unifies text-to-SQL assessment with agentic behavior evaluation into a production-grade pipeline calibrated against human expert judgment. We present BADGER, developed at Merkle, a unified evaluation framework integrating text-to-SQL assessment with agentic behavior evaluation. BADGER offers three contributions. First, LLM-assisted SQL component extraction extending Spider methodology to handle CTE-heavy, dialect-specific SQL. Second, a hybrid execution accuracy metric (Hybrid-EX) resolving column-aliasing and numeric-tolerance brittleness by using an LLM to infer structural alignments before deterministic cell-level scoring. Validated on 150 human-annotated industry queries, Hybrid-EX achieves Cohen's kappa=0.717 [95% CI: 0.600-0.822] (Substantial agreement) and 87.3% balanced accuracy, outperforming all six competing frameworks (Delta-kappa: 0.322-0.502, all p<=0.001). Third, an enterprise agentic evaluation suite assembling RAGAS, G-Eval, and agent benchmark metrics into a unified pipeline; Excess Tool Usage is the sole novel element. BADGER runs entirely within the client's governed data environment, supports configurable LLM judge backends, and enables rapid prototyping of client-specific judges and metrics, serving as a continuous evaluation backbone rather than a one-time quality gate.

Shannon Serrao, Soumitra Chatterjee, Dorina Strori +2

Merkle Analytics

LLM Agent Evaluation AI Agent Evaluation Text-to-SQL Automated Evaluation

Jun 1, 2026cs.AI

SMH-Bench: Benchmarking LLM Agents for Environment-Grounded Reasoning and Action in Smart Homes

Smart homes are evolving toward complex state-dependent living environments, requiring Large Language Models (LLMs) to reason over user intent, preferences, and multi-device interactions. However, existing smart-home benchmarks often focus on static instruction-to-API mapping or limited simulations, failing to evaluate whether LLMs can reason, interact, and act reliably in realistic household scenarios. To address these limitations, we introduce SMH-Bench, a comprehensive benchmark for evaluating LLMs in smart-home environments. Built upon HomeEnv, an executable and verifiable smart-home simulator, SMH-Bench contains 1,100 high-quality tasks spanning 7 categories and 22 fine-grained subcategories. It further stratifies tasks across simple, medium and complex homes, ranging from small apartments to dense multi-room environments with 135 devices. Experiments show that although frontier LLMs achieve strong performance on explicit control and query tasks, they still exhibit significant weaknesses in automation task scheduling, ambiguity handling and personalized reasoning, especially as home complexity increases. We hope SMH-Bench will facilitate the development of more reliable, context-aware, and practically deployable smart-home agents.

Kuan Li, Shuo Zhang, Huacan Wang +12

Midea Group · Beijing University of Posts and Telecommunications · Donghua University +2

LLM Agent Evaluation Internet of Things AI Agent Benchmarks

Jun 1, 2026cs.CL

CRAB-Bench: Evaluating LLM Agents under Complex Task Dependencies and Human-aligned User Simulation

Evaluating LLM agents in realistic service scenarios requires complex task dependencies, imperfect user behavior, and an evaluation that accommodates multiple valid solutions. We introduce CRAB-Bench (Constraint-based Realistic Agent Benchmark) and RUSE (Realistic User Simulation Engine) to address this gap. CRAB-Bench generates tasks via a constraint graph over multiple interdependent entities with structured distractors, requiring agents to reason carefully over thousands of misleading candidates where only a tiny fraction of solutions are valid. RUSE replaces cooperative, template-like simulators with realistic users grounded in human behavioral studies, instantiated across diverse personas and four behavioral dimensions. Experiments on four frontier LLM agents show that the best model achieves only 61% pass@1 on CRAB-Bench, and switching to RUSE causes further drops of up to 57%, concentrated in task-solving ability rather than conversational quality. Information Disclosure is the most damaging behavioral dimension, and agents interacting with RUSE are less likely to admit mistakes, instead masking errors through implicit corrections.

Danqing Wang, Akshay Sivaraman, Lei Li

Carnegie Mellon University

Human Behavior Simulation LLM Agent Evaluation AI Agent Benchmarks User Simulation

Jun 1, 2026cs.AI

Characterization of Multi-Model Agentic AI Systems on General Tasks via Trace-Driven Simulation

Agentic AI completes tasks through iterative planning, tool use, and reasoning based on observed outcomes. Despite its popularity, its system-level behavior remains poorly understood, particularly for complex datasets and agent architectures-owing to highly non-deterministic execution, prohibitive evaluation costs, and limited visibility into proprietary models. This paper presents GAIATrace, the first token-level trace dataset of two state-of-the-art agentic systems (MiroThinker and OWL) running GAIA, a benchmark composed of a heterogeneous mix of general-purpose tasks. Unlike prior trace datasets, GAIATrace captures full reasoning tokens, task-level structures, and activities of every major participating LLMs, enabling in-depth systems research. Complementing the dataset, we present Vidur-Agent, a trace-driven simulator that can replay GAIATrace to perform reproducible, low-cost system evaluation across diverse simulated environments. Using both artifacts, we characterize how modern agentic systems handle general tasks and how various system design choices shape their behavior, yielding several unique findings.

Donghwan Kim, Prakhar Singh, Younghoon Min +3

The Pennsylvania State University · SK Hynix · KAIST

LLM Agent Evaluation AI Agent Benchmarks Agentic AI

Jun 1, 2026cs.LG

From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents

LLM agents can fail silently by asserting task completion when the environment state shows otherwise. We study this failure mode, false success, across two agent benchmarks: 9,876 tau2-bench trajectories from 8 model families and 1,879 AppWorld trajectories from 4 model families with text-independent ground truth. False success is common but varies by setting: 45--48% of failures in single-control tau2-bench domains, 3% in dual-control telecom, and 75.8% among AppWorld self-assessing coding-agent trajectories with explicit status claims. LLM judges fail reliably: no configuration across 5 judges, 5 prompt strategies, and full task specifications exceeds AUROC 0.65 on tau2-bench, and the same judges reach only 0.54 AUROC on AppWorld API-call traces. Judges rely on surface completion proxies -- confident closing language in tau2-bench and coarse action-sequence volume in AppWorld -- rather than verified state changes. Lightweight TF-IDF detectors achieve task-disjoint AUROC 0.83 on tau2-bench and 0.95 on AppWorld, recovering 4--8x more false successes than the best judge at the same flag rate with 3,300x lower latency. These results suggest that production monitoring should use lightweight, domain-calibrated detectors as triage signals rather than relying on LLM judges as the primary monitor for false success.

Laksh Advani

University of Colorado

Agent Failure Analysis AI Agent Reliability LLM Agent Evaluation AI Agent Monitoring

May 31, 2026cs.CL

TimeSage-MT: A Multi-Turn Benchmark for Evaluating Agentic Time Series Reasoning

Time series data inform critical decisions across many real-world domains. While large language model (LLM) agents can analyze data through natural language and tools, it remains unclear whether they can conduct reliable time series analysis across multi-turn conversations. Existing benchmarks focus on single-step tasks such as forecasting and anomaly detection, overlooking practical workflows where user goals evolve, agents must build on prior analyses, and conclusions emerge from accumulated evidence. In this work, we introduce TimeSage-MT, a multi-turn benchmark for agentic time series reasoning with 240 tasks and 2,680 dialogue turns across 8 real-world domains, spanning basic exploration to decision-oriented analysis. TimeSage-MT is built through a reproducible pipeline that converts real-world time series data into multi-turn conversations with verifiable answers. It provides a unified evaluation protocol and public leaderboard for comparing time series agentic systems. To demonstrate the benchmark's utility, we evaluate frontier LLMs alongside TimeSage, a novel structured agent equipped with a comprehensive time series skill library. The results show sharp performance drops on decision-oriented tasks, driven by failures in memory, uncertainty handling, and domain-based decision making. TimeSage-MT exposes critical gaps in current agentic reasoning and provides a rigorous foundation for future development.

Yaxuan Kong, Qingren Yao, Yuqi Nie +7

University of Oxford · VulpiVox Intelligence · Eindhoven University of Technology +3

Time Series Reasoning LLM Agent Evaluation Agent Evaluation AI Agent Benchmarks

May 31, 2026cs.AI

Early Diagnosis of Wasted Computation in Multi-Agent LLM Systems via Failure-Aware Observability

Failure-aware observability diagnoses wasted computation in multi-agent LLM systems before final-answer evaluation can explain what went wrong. We propose a trace-based framework for a three-agent architecture -- orchestrator, search agent, and execution agent -- that converts structured events into online signals for loops, budget pressure, low information gain, and tool instability, then adds offline semantic grounding metrics and selective LLM-as-judge evaluation. On 165 GAIA validation traces under identical caps, 98 runs produce usable final answers and 67 fail or stop without one. Among warned failed runs, 58.1% of tokens are spent after the first warning on average, indicating substantial opportunity for intervention. A 10-task Level-2 pilot uses warnings to diversify search or require evidence, reducing post-warning token fraction from 0.638 in the baseline to 0.304. The results support a layered design: cheap online signals help the orchestrator redirect or halt redundant behavior, while deeper semantic checks identify whether completed answers are grounded enough to trust.

Xianyou Li, Weiran Yan, Yichao Wu +4

New York University · New York, NY 10012, USA · Independent Researcher +7

LLM Inference Efficiency Multi-Agent LLM Systems LLM Agent Evaluation AI Agent Monitoring

May 31, 2026cs.CR

A New Framework for Cybersecurity Refusals in AI Agents

Agentic scaffolds have dramatically improved LLM performance on complex, long-horizon tasks, yielding both broad benefits and amplified risks in domains like cybersecurity. Existing benchmarks for AI agents in cybersecurity focus mainly on measuring proficiency--how effectively agents can complete offensive security tasks--but neglect a critical question: when and how should agents refuse harmful requests? We present the first framework for establishing refusal boundaries in offensive security contexts. Our framework defines (1) principled criteria for when tasks should be refused, (2) categories of tasks that warrant refusal, and (3) evaluation methodology for measuring agent robustness under both benign and adversarial conditions. We apply this framework to assess how current LLM-powered agents adhere to appropriate refusal boundaries across a range of web-based offensive security scenarios, finding that 6 of 8 frontier models tested show near-zero refusal rates, with only 2 models (GPT-5.2 and GPT-5.1 Codex) demonstrating any meaningful refusal behavior.

Eliot Krzysztof Jones, Mateusz Dziemian, Matt Fredrikson +1

Work done while at Gray Swan · Gray Swan AI · Carnegie Mellon University

LLM Agent Evaluation LLM Refusal Behavior LLMs for Cybersecurity LLM Agent Safety

May 31, 2026cs.AI

TravelEval: A Comprehensive Benchmarking Framework for Evaluating LLM-Powered Travel Planning Agents

The development of Large Language Models (LLMs) has significantly improved travel planning applications, yet evaluating such models is limited by existing benchmarks' limitations: 1) overemphasis on constraint compliance, neglecting multi-dimensional qualities like spatio-temporal cost; 2) datasets lacking real-world authenticity and coverage in key areas (e.g., lodging, transport); and 3) isolated daily plan assessments that miss critical details (e.g., the impact of daily accommodation and visit pacing) needed for entire plan's evaluation. To address this gap, we introduce TravelEval, a realistic and comprehensive benchmark. TravelEval features 1) a novel six-dimensional evaluation framework to holistically assess plans across accuracy, compliance, temporality, spatiality, economy, and utility dimensions; 2) a highly realistic data sandbox with precise accommodation pricing and authentic intercity transportation data; and 3) a simulation-based global evaluation method that emulates complete travel plans with API-integrated geographic information and fine-grained queuing time. Evaluating 12 mainstream approaches with TravelEval reveals several valuable insights, such that LLMs struggle with globally-optimized multi-dimensional planning (especially in spatio-temporal reasoning and budget compliance), and agentic reasoning strategies offer no consistent improvement. Concisely, TravelEval facilitates travel plan evaluation via grounded spatio-temporal emulation and comprehensive metrics, providing a robust foundation for advancing LLM-powered travel planning research and applications.

Weiyi Chen, Shuaixiong Wang, Ziyun Gao +5

Zhejiang University · Hangzhou, China · Hong Kong Polytechnic University +5

Spatiotemporal Reasoning LLM Agent Evaluation AI Agent Benchmarks

May 30, 2026cs.AI

Adversarial Feeds Steer LLM Agent Decisions Against Their Defaults

LLM agents increasingly act after consuming ranked external information streams such as social feeds, search results, retrieval contexts, and email queues, yet safety evaluations almost always test the model or the user prompt in isolation, never the upstream ranker that decides what the agent reads just before it acts. We introduce a controlled protocol that holds the model, persona, topic, and final decision prompt fixed and varies only the composition and ordering of the posts an agent encounters during a preceding ten-turn "scrolling" phase, isolating the causal effect of feed curation on a downstream decision. Across 2,785 decision rollouts on four modern open instruct LLMs from three independent labs, we identify three response regimes: adversarial capitulation, default saturation, and a default-direction asymmetry in which a one-sided feed tips a decision the model was genuinely uncertain about (in the clearest cases from 5% to 100%; Fisher p as low as 3 x 10^-10) but cannot dislodge one it already favors or holds firmly. The effect follows a dose-response curve, survives a generator swap that rules out a writing-style artifact, generalizes across several decision domains including security-relevant choices such as removing a deployment approval gate or relaxing access controls, and is partly mitigated by two simple feed-level defenses; a frontier model retains its default. We characterize the recommender as a practical, default-bounded control surface for LLM agents, and argue that agent evaluations must audit the feed layer rather than the final prompt alone.

Rana Muhammad Usman

Independent Researcher

Adversarial Attacks on LLMs LLM Agent Evaluation LLM Decision-Making LLM Agent Safety

May 30, 2026cs.CL

Not All Flips Are Conformity: Decomposing Stance Convergence in Multi-Agent LLM Debate

Multi-agent debate (MAD) is a promising strategy for improving LLM reasoning, but when agents converge on a shared answer, it is unclear whether that convergence reflects genuine deliberation or social compliance. We show that the conventional answer flip rate conflates three distinct mechanisms: spontaneous instability, stance-induced conformity, and reasoning-induced persuasion. Our three-source decomposition framework isolates each through controlled counterfactual conditions. In the primary MMLU-Pro setting, 37% of agent-question observations change under self-reflection alone, while robustness tests show substantial model-dependent instability across GPQA-Diamond and three model families; strict conformity is 29% in the primary setting and remains predominantly harmful across model replications (57-77% correct-to-wrong). A controlled information-gradient experiment reveals that even vacuous reasoning is associated with 20-39% error adoption among resistant agents, with reasoning-like presentation carrying substantial persuasive weight. Harmful conformity can be predicted from Round 0 features (AUC = 0.79), and risk-targeted intervention reduces it by 13.6 percentage points (p < 0.001). However, without correctness labels or self-reflection controls, reducing peer adoption does not improve accuracy, because harmful and beneficial influence cannot be distinguished.

Xiqi Hao, Zengqing Wu, Yu-Xuan Qiu +4

Beijing Institute of Technology, Zhuhai · University of Osaka · Shenzhen University

LLM Agent Evaluation Multi-Agent Debate Multi-Agent LLMs

May 30, 2026cs.AI

ForeSci: Evaluating LLM Agents for Forward-Looking AI Research Judgment

AI research often requires decisions before future evidence exists: which bottleneck to attack, which direction to pursue, or where a project should be positioned. We introduce ForeSci, a temporally controlled benchmark for evaluating whether LLM agents can make such forward-looking research judgements from historical evidence. ForeSci contains 500 tasks across four fast-moving AI domains and four decision families. Each task is paired with a cutoff-aligned offline knowledge base; post-cutoff papers are hidden during generation and used only for validation. To avoid random future-event prediction, tasks are derived from pre-cutoff taxonomy branches and evidence signals, and answer-generation backbones are selected to precede the task cutoffs. We evaluate native LLMs, Hybrid RAG, and three research-agent adaptations across four backbones. Results show that explicit evidence organization improves traceability and factual support, but gains depend strongly on the decision family. Diagnostics reveal a recurring evidence-decision decoupling: agents may cite relevant evidence while forecasting the wrong research object. ForeSci turns forward-looking AI research judgement into a controlled benchmark for evaluating research agents as decision-making systems.

Qiuyu Tian, Haojie Yin, Yingce Xia +2

1Southeast University, Nanjing, China · 3Duke Kunshan University, Kunshan, China · 2Beijing Zhongguancun Academy, Beijing, China

LLM Agent Evaluation AI Agent Benchmarks LLM Decision-Making

May 30, 2026cs.AI

Doing What They Say, Not What They Reason: Locating the Faithfulness Gap in LLM Agents

Do LLM agents act on the reasoning they state? This question of process fidelity is central to LLM-based social simulation, yet hard to measure where no reference for correct behavior exists. We study it in a controlled setting: a Texas Poker simulator with a verifiable reference action for every decision by splitting the faithfulness gap into two steps: reasoning-to-conclusion (does the stated decision follow from the agent's own reasoning?) and conclusion-to-action (does the agent execute what it states?). The two steps behave very differently. Conclusion-to-action is reliable: inconsistency is 0.7% for Claude Haiku 4.5 and 1.4% for DeepSeek-Reasoner once the conclusion is read from an explicit tag, whereas free-text conclusion extraction reports 22-26%. Reasoning-to-conclusion is where fidelity frays, but not through a single dominant failure. In a step-level diagnostic the agent's errors split roughly evenly between bad inputs, borderline cases, and rule misapplication deriving a conclusion that contradicts the agent's own restated rule from inputs it estimated correctly. This composition is model-dependent: rule misapplication accounts for a third of Haiku's interpretable errors but only 8% of DeepSeek's. The one robust signal is directional: when an agent does misapply its own stated rule, it almost always (99.5% for Haiku) errs in the risk-averse direction. The override is partly hedging behavior, not a capability limit: instructing the agent to apply the rule mechanically halves the misapplication rate (13.9% to 6.8% of decisions) and raises adherence by eight points. Process-fidelity evaluation should therefore elicit machine-checkable conclusions and probe for directional biases rather than assume a single upstream failure mode, lest it conflate measurement noise with model behavior.

Yufeng Wang

AI Agent Reliability LLM Agent Evaluation LLM Decision-Making Agentic Reasoning LLM Reasoning

May 29, 2026cs.LG

BAGEN: Are LLM Agents Budget-Aware?

While agents are increasingly spending more resources, today agent cost is mostly measured only after execution. A Budget-Aware Agent (BAGEN) should treat budget as an active control signal, rather than a passive cost metric. We first systematically define budget estimation as internal budgets (from agent computation) and external budgets (from agent actions). We then formalize budget-awareness as progressive interval estimation: at each step of a plan, an agent should predict an upper and lower bound on remaining budget, and alert when completion is unlikely. Scoring with a rollout-replay protocol, we find consistent failure patterns on four environments and five frontier agents: (1) strong agents do not necessarily have strong budget-awareness, with correlation r=0.35. (2) frontier models are consistently over-optimistic, continue spending on tasks that are unlikely to succeed, instead of alerting the user early. (3) budget-aware signal is actionable and trainable. Early stop saves 28-64% tokens on failed trajectories, and SFT+RL strengthens early stop and alert behavior. (4) precise interval calibration remains challenging, with interval coverage capping at 47% after SFT+RL. Project page: https://ragen-ai.github.io/bagen/

Yuxiang Lin, Zihan Wang, Mengyang Liu +9

1Northwestern University · 2O2 Lab · 3Independent +5

LLM Agent Evaluation AI Agent Benchmarks LLM Agents LLM Agent Training

May 29, 2026cs.CL

Skill Availability and Presentation Granularity in Large-Language-Model Agents: A Controlled SkillsBench Study

Skill documents provide procedural knowledge to large-language-model agents at inference time. This article studies whether the presentation granularity of controlled skill knowledge changes downstream task success. The experiment uses a pinned SkillsBench version, a 30-task domain-balanced subset validated by official oracle runs, two reasoning-enabled model configurations, six skill conditions, and five trials per task-condition-model cell. Skill availability is the clearest empirical signal. Relative to no skill, skill conditions increase task-mean pass rate by 26.7 to 36.0 percentage points for GPT-5.5 and by 18.0 to 26.0 percentage points for DeepSeek V4-Flash. The final data contain 1,800 rows, with 900 rows for each model. The task is the inference unit. Five trials are aggregated within each task-condition-model cell before paired contrasts are estimated over 30 tasks. The primary presentation contrasts are smaller and uncertain. Low-abstraction guidance differs from high-abstraction guidance by +0.7 percentage points for GPT-5.5 and -6.7 percentage points for DeepSeek V4-Flash, with both 95% bootstrap confidence intervals crossing zero. Adding one worked example to medium-abstraction guidance differs from the no-example variant by +0.7 and +1.3 percentage points. Mean-reward robustness checks preserve the same substantive conclusion. In this controlled subset, skill availability is associated with higher success than no skill, while the tested presentation-granularity changes yield small, uncertain, and model-dependent effects.

Xiaonan Xu, Wenjing Wu

Computer Information Technology, Northern Arizona University, Flagstaff, AZ, USA · Computer Science, University of Colorado Boulder, Boulder, CO, USA

LLM Agent Evaluation LLM Agent Skill Learning AI Agent Benchmarks LLM Agents

May 29, 2026cs.AI

Industrializing Prediction-Powered Inference: The GLIDE Library for Reliable GenAI and Agentic Systems Evaluation

Reliable evaluation of agentic systems requires unbiased estimates with valid uncertainty, but standard practice navigates between costly human annotation and biased LLM-as-judge proxies. Prediction-powered inference (PPI) combines both into debiased estimates with valid confidence intervals, yet its various methods remain scattered across papers under partial implementations. We introduce GLIDE, an open-source Python library that unifies state-of-the-art PPI estimators (PPI++, Stratified PPI, Predict-Then-Debias and its stratified variants, Active Statistical Inference) and samplers (uniform, stratified, active, cost-optimal) under a scipy-style API specialized to mean estimation. GLIDE ships with a reproducible Monte Carlo validation suite, an empirically grounded decision tree for method selection, and an agentic evaluation case study showing substantial annotation savings at equivalent precision. The GLIDE package is available at this URL: https://github.com/EmertonData/glide

Grégoire Martinon, Ibrahim Merad, Mohammed Raki

Emerton Data, 16 Avenue Hoche, 75008, Paris, France.

LLM-Assisted Annotation LLM Agent Evaluation Prediction-Powered Inference Generative AI Evaluation

May 29, 2026cs.HC

TUX: Measuring Human--AI Tacit Understanding

As large language models (LLMs) increasingly act as collaborative partners, human--AI alignment is often evaluated through explicit task success, accuracy, or reward optimization. Yet many collaborative settings depend on tacit understanding: whether an agent can align with a human's evaluative stance or representational priors without clear objectives, communication, or feedback. To study this capacity, we develop a spectrum-placement task inspired by the social party game Wavelength, in which humans and agents independently place concepts along subjective spectra. We operationalize the Tacit Understanding Index (TUX) as a pairwise behavioral measure of similarity between human and agent judgments, and evaluate it with 241 human participants and 200 profile-conditioned LLM agents across four models. We find that nearest human--agent pairs in trait space achieve significantly higher TUX, suggesting that tacit alignment is associated with person-level characteristics rather than reflecting only random similarity. Regression analyses show that TUX becomes more explainable as predictor sets become richer, with individual traits, decision-making styles, and confidence improving over aggregate trait-distance baselines. These findings suggest that TUX provides a measurable behavioral signal of human--LLM tacit understanding, while revealing the limits of profile-based conditioning for capturing deeper representational alignment.

Yueshen Li, Hanyi Min, Vedant Das Swain +1

University of Illinois Urbana-Champaign · New York University

LLM Alignment LLM Agent Evaluation Human-AI Interaction Human-AI Collaboration

May 29, 2026cs.SE

BlueFin: Benchmarking LLM Agents on Financial Spreadsheets

We present BlueFin, a benchmark that tasks large language model (LLM) agents with synthesis, manipulation, and comprehension tasks over spreadsheet workbooks in the professional finance domain. Though estimates of the global population of paying users of spreadsheet software range in the hundreds of millions -- an order of magnitude more than the estimated global population of professional developers -- comparatively fewer resources have been devoted to exploring and expanding LLM capabilities in the spreadsheet domain, with fewer still dedicated to mirroring real occupational tasks encountered by those in professional finance roles. In response, we curate a set of 131 challenging, complex tasks with real-world relevance in the domain, containing 3,225 granular rubric criteria; notably, our rubric criteria and LM judge evaluations are validated by a team of expert human annotators, resulting in high-quality, granular evaluations of complex tasks that are difficult to verify programmatically but can be reliably evaluated by an LM judge agent. Our judge achieves parity with expert consensus (

α=0.826

) with a macro-F1 score of 0.839. Frontier LLMs demonstrate poor performance on the challenging benchmark, with the strongest LLMs achieving less than 50% average scores across tasks -- models exhibit particular weaknesses in dynamic correctness. Our contributions include a dataset of examples across three categories of spreadsheet tasks, an open source harness and agentic evaluation framework, and a characterization of existing frontier models' performance on our benchmark.

Srivatsa Kundurthy, Clara Na, Colton Moraine +6

1Longitude Labs Inc. · 2Cornell University · 3Carnegie Mellon University

LLM Agent Evaluation AI Agent Benchmarks Spreadsheet Automation Finance

May 29, 2026cs.AI

MAVEN: Improving Generalization in Agentic Tool Calling

Generalization across agentic tool-calling environments remains a central challenge for reliable agentic reasoning systems. Although large language models achieve strong results on individual benchmarks, their ability to compose reasoning strategies, preserve intermediate states, and coordinate tools across domains remains underexplored. We present MAVEN (Modular Agentic Verification and Execution Network), a lightweight symbolic reasoning scaffold for structured decomposition, adaptive tool orchestration, and intermediate verification. We evaluate MAVEN across established tool-calling benchmarks, including BFCL v3, TauBench, Tau2Bench, AceBench, and introduce MAVEN-Bench, a stress-test benchmark for multi-step mathematical and physical reasoning with explicit verification and adversarial task composition. MAVEN-Bench exposes a substantial gap between partial reasoning quality and end-to-end task success; in direct MAVEN-Bench runs, MAVEN improves its GPT-OSS-120b base model from 48% to 71% accuracy without additional training. It also remains competitive with frontier proprietary baselines while using an open-weight backbone with an estimated cost ratio of roughly 1/10, suggesting that lightweight verification-centered scaffolds can strengthen compositional reasoning and motivate more process-aware evaluation of agents in the wild.

Omkar Ghugarkar, Vishvesh Bhat, Muhammad Ahmed Mohsin +1

CoreThink AI, USA · Stanford University, Stanford, CA, USA

Logical Reasoning LLM Agent Evaluation Symbolic Reasoning AI Agent Benchmarks Agentic Reasoning LLM Tool Use+1

May 29, 2026cs.CR

Depth-Dependent Indirect Prompt Injection in Tool-Calling ReAct Agents: Injection Depth, Payload Framing, and Turn-Budget Sensitivity

ReAct agents that interleave chain-of-thought reasoning with tool calls are increasingly deployed for real tasks such as scheduling, file retrieval, and data access. Their tool observation loop creates a direct attack surface: an adversary who controls any tool's return value can embed instructions that redirect the agent away from the user's goal, a threat known as indirect prompt injection. Existing benchmarks evaluate attack success rate (ASR) at a fixed injection position under fixed conditions, leaving three risk dimensions unexplored: where in the tool sequence the payload appears (injection depth), what rhetorical register it uses (framing), and how many turns the agent is permitted (turn cap). We conduct four controlled studies on 20 scenarios spanning five attack categories, totalling 460 trials against GPT-4o-mini and Claude Haiku at a combined API cost under 0.36 USD. Study 1 shows that ASR against GPT-4o-mini decays from 60% at depth 1 to 0% at depths 4 and 5 (Cramer's V = 0.58, p < 0.001; restricted to within-sequence depths 1-3: V = 0.47, p = 0.0013), driven by model resistance at depth 1 and task completion before payload encounter at deeper positions. Study 2 replicates the depth experiment on Claude Haiku, which achieves 0% ASR at every depth through a combination of conservative tool invocation and genuine instruction resistance. Study 3 shows that framing modulates ASR between 25% (neutral) and 75% (persona) at depth 1, a 50-percentage-point range that does not reach statistical significance at N = 20 per condition. Study 4 confirms that ASR is stable across turn caps of 3, 5, and 7, indicating the turn budget is not a risk factor in this setting. Our results establish injection depth as the dominant variable and show that sanitising only the first tool observation captures 67% of measured injection successes.

Mohammadreza Rashidi

Department of Computer Science AI and Media Analysis Lab Berlin, Germany

LLM Agent Evaluation AI Agent Security Indirect Prompt Injection Prompt Injection Attacks Prompt Injection Attacks on LLMs Prompt Injection Attacks on AI Agents

May 28, 2026cs.LG

On Effectiveness and Efficiency of Agentic Tool-calling and RL Training

Tool-calling is a central component of modern large language model (LLM) agents, equipping them with skills beyond their parametric knowledge. This paper studies tool-calling along two complementary axes: effectiveness, i.e., how this capability is measured, and efficiency, i.e., how it is learned. On effectiveness, we systematically analyze tool-calling evaluation pipelines and show that results can be highly sensitive to seemingly minor, often undocumented implementation choices including the random seed, system prompt, multi-turn template construction, and how prior interaction/reasoning history is carried forward. These choices can lead to substantial differences in reported performance, especially in multi-turn settings where without rigorous standardization, leaderboard rankings are unreliable. On efficiency, we examine standard reinforcement learning (RL) for tool-calling and identify two sources of computational waste: (i) during rollouts, many prompts produce no learning signal, and (ii) during policy updates, optimization incurs high computational cost. Guided by these findings, we introduce two techniques that accelerate RL-based tool-calling training, achieving substantial wall-clock speedup without degrading performance.

Tong Liu, Cheng Qian, Matej Cief +4

1LMU Munich & Munich Center for Machine Learning · 2UIUC · 3Amazon +1

Tool-Use Evaluation LLM Agent Evaluation RL for Tool Use

May 28, 2026cs.AI

Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents

LLM agents are increasingly deployed as systems built around editable external harnesses, including prompts, skills, memories and tools, that shape task execution without changing model parameters. Harness self-evolution adapts such agents by updating these harnesses from execution evidence. Yet it remains unclear whether a model's base capability in task-solving predicts its capabilities in harness self-evolution: which models produce useful harness updates, and which actually benefit from them? We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving. Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains; even Qwen3.5-9B's updates yield gains comparable to those of Claude Opus~4.6. Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier. We trace low gains at the weak tier to two failure modes: weak-tier models may fail to activate relevant harness artifacts, or activate them but fail to follow them faithfully. These findings suggest investing capability budget in the task-solving agent rather than the evolver, and targeting harness invocation and long-horizon instruction following in agent training. Our source code is publicly available at https://github.com/A-EVO-Lab/a-evolve/tree/release/harness-evolution.

Minhua Lin, Juncheng Wu, Zijun Wang +14

1The Pennsylvania State University · 2UC Santa Cruz · 3Amazon +3

LLM Agent Evaluation Agent Harness Optimization LLM Agent Self-Improvement LLM Agent Harnesses Self-Evolving Agents

May 28, 2026cs.LG

Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents

Two clinical AI systems can score nearly identically on coverage-based rubrics yet behave radically differently when their patient inputs change: one updates its recommendations to match the new clinical signal, while the other produces the same output regardless. We introduce the Causal Sensitivity Score (CSS), a pre-registered interventional metric that mutates oncology tumor-board cases along five clinically meaningful dimensions - biomarker flips, prior-treatment failures, biomarker removals, surgery-status changes, and stage perturbations - and scores whether each model updates its recommendations in the pre-registered correct direction using a {0, 0.5, 1.0} scale. Benchmarked against the Consensus Match Score (CMS), a coverage-based weighted recall metric, six frontier models from three labs evaluated in single-shot inference across 224 cases rank in nearly opposite orders: all six models change rank, the CMS-worst model becomes CSS-best, and one upper-mid CMS model ranks last on CSS. We further surface a universal safety blind spot: every frontier model fails on surgery-status interventions (at most 17.2% CSS on Family D), a finding CMS does not expose. The metric also transfers to tool-using agents: in a ReAct-style experiment, tool use improves CSS for five of six models (+2.5 to +20.3 percentage points), yet the lowest-CSS model retrieves the same chart sections and still fails to update its recommendations - revealing a structural responsiveness deficit visible only under counterfactual evaluation. Cross-judge replication and three-rater medical-professional validation confirm the aggregate findings. Interventional pre-registered metrics like CSS complement coverage-based evaluation for clinical AI agents: they capture responsiveness that coverage metrics miss and offer a candidate dense reward signal for future agentic RL systems.

Matt Turk

Protege Data Lab New York City, USA

Counterfactual Evaluation Healthcare Clinical Decision Support LLM Agent Evaluation Clinical Language Model Evaluation

May 28, 2026cs.CL

Can LLM Teams Play What? Where? When?

Large language models (LLMs) remain limited on tasks requiring indirect reasoning, cultural knowledge, and coordinated hypothesis testing. We investigate whether team-based interaction improves LLM performance in What? Where? When? (ChGK), a quiz game designed to reward collective reasoning. We introduce three team strategies: Voting, Silent Team (the captain observes final answers), and Talkative Team (the captain observes both answers and rationales). To minimize data leakage, we evaluate these strategies on a dataset consisting of 572 ChGK questions released in 2025. Using six recent large-scale open models, we show that team-based strategies outperform single-model baselines, yielding gains of up to 20 percentage points in accuracy. The best team achieves 44.23% accuracy, and approaches human team performance on questions with available human statistics. Analysis of inter-model diversity reveals that disagreement strongly predicts lower accuracy, but explanatory communication substantially mitigates performance drops. We further examine captain behavior and find no evidence of self-preference bias; access to peer rationales improves captain judgments. Overall, LLM teams function primarily as answer selection and error-filtering mechanisms rather than generators of novel solutions. Our findings highlight the importance of interaction and suggest adaptive strategies as a promising direction for multi-agent systems.

Anastasia Kotelnikova, Viktor Byzov, Maria Dolzhenkova +1

Vyatka State University Kirov, Russia · European University at St. Petersburg St. Petersburg, Russia

Multi-Agent LLM Systems LLM Agent Evaluation Question Answering Multi-Agent Reasoning Multi-LLM Collaboration

May 28, 2026cs.MA

SpecBench: Evaluating Specification-Level Reasoning for Software Engineering LLM Agents

Software engineering (SWE) agents are transitioning from code generation to full software development lifecycle automation. A critical phase in this lifecycle is specification design: transforming initial proposals into carefully considered requirements through expert review. Existing benchmarks such as SWE-Bench are implementation-focused by measuring the agent's ability to generate code given fixed, precise design requirements. This formulation assumes specifications are correct and complete. In real-world complex and critical software systems, initial specifications are often incomplete and flawed, requiring extensive expert reviews and revisions before being accepted for implementation. To fill this gap, we introduce SpecBench to evaluate specification-level reasoning: the ability to generate complete, unambiguous, consistent, and correct system specifications. SpecBench tasks are derived from the Request for Comments (RFC) process used by mature open-source projects. For each task, an agent is given an initial design proposal, the project codebase, and all past project RFC discussions. The agent is tasked with identifying specification deficiencies: omissions, ambiguities, inconsistencies, or incorrect assumptions in the initial proposal. We evaluate predictions against critiques raised by expert maintainers during historical RFC reviews. SpecBench contains tasks from 5 diverse repositories: Kubernetes, React, Rust, TVM, and vLLM. We evaluate state-of-the-art SWE agents on SpecBench, analyzing their capacity to reason about system design without execution feedback. The best performing agent, GPT-5.4, achieves 44.4% accuracy.

Grant Hamblin, Kevin Song, Zhanda Zhu +4

University of Toronto · Toronto, Ontario, Canada · University of Waterloo +3

Requirements Engineering Software Engineering Agents Software Engineering Benchmarks LLM Agent Evaluation

May 28, 2026cs.MA

Specialize Roles, Mix Deployments: Pushing the Cost-Accuracy Frontier of LLM Agent Teams

LLM agents are increasingly deployed as multi-role teams, where tasks are divided across specialized roles such as planner, executor, and verifier. In these systems, cost and accuracy are no longer properties of a single model: they depend on which model fills each role and where it is hosted, including API, self-hosted, and hybrid deployment. Existing agentic benchmarks typically evaluate fixed models or fixed agent configurations, and therefore offer limited guidance for cost-accuracy-optimal deployment. We introduce AgentCARD, a role-aware benchmark suite for evaluating LLM agent teams across role assignment and deployment mode. AgentCARD combines a role-decomposed evaluation harness, a unified API/self-hosted cost model, Pareto-frontier analysis, and a Shapley-based diagnostic for identifying role bottlenecks. Our evaluation shows that heterogeneous teams consistently occupy the cost-accuracy frontier. They improve accuracy by up to

44\%

over cost-equivalent homogeneous teams, or match the strongest homogeneous team at up to

12\times

lower per-task cost through hybrid deployment. We further find that the best role assignment is domain-dependent: some domains are planner-bottlenecked, while others are executor-bottlenecked. Finally, AgentCARD extends beyond planner--executor teams to workflows with additional roles such as verification, and supports continual evaluation as new domains and team structures emerge. Our code is released at: https://github.com/Auto-CAP/AgentCAP

Yinsicheng Jiang, Liang Cheng, Yeqi Huang +6

University of Edinburgh · 2Microsoft Research

Multi-Agent LLM Systems LLM Agent Evaluation LLM Agent Workflow Optimization AI Agent Benchmarks LLM Agents

May 28, 2026cs.LG

ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research

AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify. We present ResearchClawBench, a benchmark for evaluating autonomous scientific research across 40 tasks from 10 scientific domains. Each task is grounded in a real published paper, provides related literature and raw data, and hides the target paper during evaluation. Expert-curated multimodal rubrics decompose the target scientific artifacts into weighted criteria, enabling evaluation of target-paper-level re-discovery while leaving room for new discovery. We evaluate seven autonomous research (auto-research) agents under a unified protocol and seventeen native LLMs through the lightweight ResearchHarness. Current systems remain far from reliable re-discovery: the strongest autonomous agent, Claude Code, averages 21.5, and the strongest ResearchHarness LLM, Claude-Opus-4.7, averages 20.7, with an LLM frontier mean of only 26.5. Error analysis shows that failures concentrate in experimental protocol mismatch, evidence mismatch, and missing scientific core. ResearchClawBench provides a reproducible evaluation frontier for measuring progress toward autonomous scientific research.

Wanghan Xu, Shuo Li, Tianlin Ye +48

LLM Agent Evaluation AI Agent Benchmarks Autonomous Scientific Discovery Autonomous Research Agents

May 28, 2026cs.CL

HEART-Bench: Do LLM Agents Exhibit Human-like Psychology?

While LLM agents have demonstrated remarkable task-oriented abilities such as planning, reasoning, and action, few works have treated them as complete human personalities where emotional dimensions hold equal importance. In this paper, we introduce a novel benchmark to systematically assess whether LLM agents can simulate coherent, human-like psychology. Specifically, our benchmark constructs 11 diverse human characters grounded in orthogonal Big Five personality traits, with each profile deeply integrated with 1,000 structured autobiographical-style episodic memories distributed across theory-grounded developmental life stages. To rigorously evaluate the psychological manifestations of LLMs, we designed a curated suite of 64 decision-making scenarios, guided by the DIAMONDS taxonomy, a psychological framework that characterizes situations along eight dimensions: Duty, Intellect, Adversity, Mating, pOsitivity, Negativity, Deception, and Sociality. By subjecting agents to varying scenarios, the benchmark evaluates whether they can consolidate their innate personality traits and autobiographical memories to make behavioral decisions that are consistent with their specific psychological profiles. After systematic human validation and filtering, we obtained a benchmark consisting of 673 multiple-choice questions (MCQs). We believe this benchmark provides a principled and scalable testbed for studying human-like emotions, personality consistency, and value-consistent behavioural decision-making in LLM-based agents.

Weihan Peng, Chenxu Zhang, Qianao Wang +7

Shanghai Jiao Tong University · Imperial College London · Quwan Group +2

Personality Modeling in Language Models Human Behavior Simulation LLM Agent Evaluation AI Agent Benchmarks

May 28, 2026cs.CL

Does The Way You Plan Matter? An Empirical Study of Planning Representations for LLM Web Agents

Despite recent advances, LLM-based web agents still struggle with limited exploration, omission of critical steps, and sensitivity to task constraints. Prior work suggests that many of these failures stem from weaknesses in planning, yet the impact of alternative natural language plan representation remains unexplored. To address this, we introduce PlanAhead, a static planner-executor framework that evaluates the impact of plan representation in agent performance. We first automatically categorize WebArena tasks into 3 difficulty levels, enabling consistent difficulty grading without human annotation. Then we systematically evaluate 4 different plan representations on the tasks categorized as hard: sequential subgoals, narrative, pseudocode, and checklist; across different families of multimodal LLM powered agents (OpenAI, Alibaba, and Google). To account for stochastic variability, we introduce two novel evaluation metrics: Achievement Rate (AR) and Solved-Task Consistency (STC). Our results show that both, the plan formulation and the underlying LLM generating the plan, significantly influence web-agent robustness and task success.

Alejandra Zambrano, Sara Vera Marjanovic, Imene Kerboua +2

Concordia University · Mila - Quebec AI Institute · University of Copenhagen +2

LLM Agent Evaluation LLM Agent Planning Web Agents

May 28, 2026cs.AI

Redundant or Necessary? A Benchmark for Detecting Redundant Steps in Agent Trajectories

LLM-based agents have demonstrated strong capabilities in solving complex tasks through multi-step reasoning and tool use. However, existing evaluation protocols primarily focus on task success, overlooking a critical aspect of agent behavior: execution efficiency. In practice, agent trajectories often contain redundant steps that consume substantial resources while contributing little to task completion. In this work, we propose and formulate a new research area: \textbf{redundant step detection} for agent trajectories. To support this initiative, we introduce \textbf{RedundancyBench}, a new benchmark that contains diverse tasks with carefully annotated trajectories, where each step is labeled according to its contribution to task completion. Using RedundancyBench, we develop and evaluate 3 representative methods to answer whether a step within trajectory is redundant or necessary. Our results show that even the best-performing method achieves only 24.88% score in detecting redundant steps, while some methods perform worse than random guessing. These results highlight the task's complexity and the need for further research in this area. \footnote{Code and dataset in this paper are both available in https://anonymous.4open.science/r/RedundancyBench.}

Minyang Hu, Bo Yang, Zhinuo Zhou +4

† Huawei Technologies, Noah Arks’ Lab · # Institute of Computing Technology, Chinese Academy of Sciences

LLM Agent Evaluation Agent Trajectory Analysis Agent Benchmarks