LLM Agent Evaluation

LLM: Large Language Model

Ask about this topic

Overview

The field studies whether language-model agents can use tools, manage state, follow instructions, and achieve users’ goals across multi-step interactions. Benchmarks place agents in executable settings such as coding repositories, websites, terminals, business workflows, and medical or scientific tasks, while methods use hidden tests, state checks, trajectory analysis, and simulated users to assess performance. Evaluation increasingly includes cost, latency, privacy, safety, consistency, and process quality—not just whether the final answer is correct. Results remain difficult to compare because harnesses and scoring choices can change rankings, and strong performance on short or isolated tasks often fails to transfer to long-horizon real-world work.

Latest papers

CardsList

1,217 papers231 papers in the last four weeks, up 62% on the four weeks before. 1.3% of all new papers.

Apr 23, 2026

When Agents Look the Same: Quantifying Distillation-Induced Similarity in Tool-Use Behaviors

Model distillation is a primary driver behind the rapid progress of LLM agents, yet it often leads to behavioral homogenization. Many emerging agents share nearly identical reasoning steps and failure modes, suggesting they may be distilled echoes of a few dominant teachers. Existing metrics, however, fail to distinguish mandatory behaviors required for task success from non-mandatory patterns that reflect a model's autonomous preferences. We propose two complementary metrics to isolate non-mandatory behavioral patterns: Response Pattern Similarity (RPS) for verbal alignment and Action Graph Similarity (AGS) for tool-use habits modeled as directed graphs. Evaluating 18 models from 8 providers on ττ-Bench and τ2τ^2-Bench against Claude Sonnet 4.5 (thinking), we find that within-family model pairs score 5.9 pp higher in AGS than cross-family pairs, and that Kimi-K2 (thinking) reaches 82.6% SnodeS_{\text{node}} and 94.7% SdepS_{\text{dep}}, exceeding Anthropic's own Opus 4.1. A controlled distillation experiment further confirms that AGS distinguishes teacher-specific convergence from general improvement. RPS and AGS capture distinct behavioral dimensions (Pearson rr = 0.491), providing complementary diagnostic signals for behavioral convergence in the agent ecosystem. Our code is available at https://github.com/Syuchin/AgentEcho.

Chenghao Yang, Yuning Zhang, Zhoufutu Wen +4

School of Cyber Science and Technology, USTC · Anhui Province Key Laboratory of Digital Security · 3M-A-P +1

Tool-Augmented Language Model AgentsLLM Agent EvaluationAI Agent BenchmarksTool-Using AgentsKnowledge DistillationBehavioral Consistency of LLM Agents
AskPDF
Apr 22, 2026

Chasing the Public Score: User Pressure and Evaluation Exploitation in Coding Agent Workflows

Frontier coding agents are increasingly used in workflows where users supervise progress primarily through repeated improvement of a public score, namely the reported score on a public evaluation file with labels in the workspace, rather than through direct inspection of the agent's intermediate outputs. We study whether multi-round user pressure to improve that score induces public score exploitation: behavior that raises the public score through shortcuts without improving hidden private evaluation. We begin with a preliminary single-script tabular classification task, where GPT-5.4 and Claude Opus 4.6 both exploit label information within 10 rounds of user-agent interaction. We then build AgentPressureBench, a 34-task machine-learning repository benchmark spanning three input modalities, and collect 1326 multi-round trajectories from 13 coding agents. On our benchmark, we observe 403 exploitative runs, spanning across all tasks. We also find that stronger models have higher exploitation rates, supported by a significant Spearman rank correlation of 0.77. Our ablation experiments show that higher user pressure leads to earlier exploitation, reducing the average first exploit round by 15.6 rounds (i.e., 19.67 to 4.08). As a mitigation, adding explicit anti-exploit wordings in prompt mostly eliminates exploitation (100% to 8.3%). We hope that our work can bring attention to more careful use of coding agents workflow, and developing more robust coding agents under user pressure. Our project page is at https://ucsc-vlaa.github.io/AgentPressureBench .

Hardy Chen, Nancy Lau, Haoqin Tu +8

UC Santa Cruz · UT Dallas · NUS

Coding AgentsLLM Agent EvaluationAI Agent Benchmarks
AskPDF
Apr 22, 2026

SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation on Real-World Tasks

Skills have become the de facto way to enable LLM agents to perform complex real-world tasks with customized instructions, workflows, and tools, but how to learn them automatically and effectively remains unclear. We introduce SkillLearnBench, the first benchmark for evaluating continual skill learning methods, comprising 20 verified, skill-dependent tasks across 15 sub-domains derived from a real-world skill taxonomy , evaluated at three levels: skill quality, execution trajectory, and task outcome. Using this benchmark, we evaluate recent continual learning techniques, those leveraging one-shot, self/teacher feedback, and skill creator to generate skills from agent experiences. We find that all continual learning methods improve over the no-skill baseline, yet consistent gains remain elusive: no method leads across all tasks and LLMs, and scaling to stronger LLMs does not reliably help. Continual learning improves tasks with clear, reusable workflows but struggles on open-ended tasks, and using stronger LLM backbones does not consistently produce better skills. Our analysis also revealed that multiple iterations in continual learning facilitate genuine improvement via external feedback, whereas self-feedback alone induces recursive drift. Our data and code are open-source at https://github.com/cxcscmu/SkillLearnBench to enable further studies of automatic skill generation and continual learning techniques.

Shanshan Zhong, Yi Lu, Jingjie Ning +7

Carnegie Mellon University · Amazon AGI

Continual Learning for LLM AgentsLLM Agent EvaluationLLM Agent Skill LearningAI Agent BenchmarksAgent Skill Learning
AskPDF
Apr 21, 2026

Time Series Augmented Generation for Financial Applications

Evaluating the reasoning capabilities of Large Language Models (LLMs) for complex, quantitative financial tasks is a critical and unsolved challenge. Standard benchmarks often fail to isolate an agent's core ability to parse queries and orchestrate computations. To address this, we introduce a novel evaluation methodology and benchmark designed to rigorously measure an LLM agent's reasoning for financial time-series analysis. We apply this methodology in a large-scale empirical study using our framework, Time Series Augmented Generation (TSAG), where an LLM agent delegates quantitative tasks to verifiable, external tools. Our benchmark, consisting of 100 financial questions, is used to compare multiple SOTA agents (e.g., GPT-4o, Llama 3, Qwen2) on metrics assessing tool selection accuracy, faithfulness, and hallucination. The results demonstrate that capable agents can achieve near-perfect tool-use accuracy with minimal hallucination, validating the tool-augmented paradigm. Our primary contribution is this evaluation framework and the corresponding empirical insights into agent performance, which we release publicly to foster standardized research on reliable financial AI.

Anton Kolonin, Alexey Glushchenko, Evgeny Bochkov +1

SingularityNET Foundation

LLM Agent EvaluationAI Agent BenchmarksLLM AgentsFinancial Time SeriesFinance
AskPDF
Apr 21, 2026

Do Agents Dream of Root Shells? Partial-Credit Evaluation of LLM Agents in Capture the Flag Challenges

Large Language Model (LLM) agents are increasingly proposed for autonomous cybersecurity tasks, but their capabilities in realistic offensive settings remain poorly understood. We present DeepRed, an open-source benchmark for evaluating LLM-based agents on realistic Capture The Flag (CTF) challenges in isolated virtualized environments. DeepRed places an agent in a Kali attacker environment with terminal tools and optional web search, connected over a private network to a target challenge, and records full execution traces for analysis. To move beyond binary solved/unsolved outcomes, we introduce a partial-credit scoring method based on challenge-specific checkpoints derived from public writeups, together with an automated summarise-then-judge labelling pipeline for assigning checkpoint completion from logs. Using DeepRed, we benchmark ten commercially accessible LLMs on ten VM-based CTF challenges spanning different challenge categories. The results indicate that current agents remain limited: the best model achieves only 35% average checkpoint completion, performing strongest on common challenge types and weakest on tasks requiring non-standard discovery and longer-horizon adaptation.

Ali Al-Kaswan, Maksim Plotnikov, Maxim Hájek +3

Delft University of Technology Delft, The Netherlands

LLM Agent EvaluationAI Agent Security BenchmarksLLMs for CybersecurityAutomated Grading
AskPDF
Apr 21, 2026

Forage V2: Knowledge Evolution and Transfer in Autonomous Agent Organizations

Autonomous agents operating in open-world tasks -- where the completion boundary is not given in advance -- face denominator blindness: they systematically underestimate the scope of the target space. Forage V1 addressed this through co-evolving evaluation (an independent Evaluator discovers what "complete" means) and method isolation (Evaluator and Planner cannot see each other's code). V2 extends the architecture from a single expedition to a learning organization: experience accumulates across runs, transfers across model capabilities, and institutional safeguards prevent knowledge degradation. We demonstrate two claims across three task types (web scraping, API queries, mathematical reasoning). Knowledge accumulation: over six runs, knowledge entries grow from 0 to 54, and denominator estimates stabilize as domain understanding deepens. Knowledge transfer: a weaker agent (Sonnet) seeded with a stronger agent's (Opus) knowledge narrows a 6.6pp coverage gap to 1.1pp, halves cost (9.40 to 5.13 USD), converges in half the rounds (mean 4.5 vs. 7.0), and three independent seeded runs arrive at exactly the same denominator estimate (266), suggesting organizational knowledge calibrates evaluation itself. V2's contribution is architectural: it designs institutions -- audit separation, contract protocols, organizational memory -- that make any agent more reliable upon entry. The accumulated experience is organizational, model-agnostic, and transferable, stored as readable documents that any future agent inherits regardless of provider or capability level.

Huaqing Xie

Independent Researcher

LLM Agent EvaluationMemory-Augmented AgentsSelf-Evolving Agents
AskPDF
Apr 21, 2026

AutomationBench

Existing AI benchmarks for software automation rarely combine cross-application coordination, autonomous API discovery, and policy adherence. Real business workflows demand all three: a single task may span a CRM, inbox, calendar, and messaging platform - requiring the agent to find the right endpoints, follow a policy document, and write correct data to each system. To address this gap, we introduce AutomationBench, a benchmark for evaluating AI agents on cross-application workflow orchestration via REST APIs. Drawing on real workflow patterns from Zapier's platform, tasks span Sales, Marketing, Operations, Support, Finance, and HR domains. Agents must discover relevant endpoints themselves, follow layered business rules, and navigate environments with irrelevant and sometimes misleading records. Grading is programmatic and end-state only: whether the correct data ended up in the right systems. Even the best frontier models currently score below 10%. AutomationBench provides a challenging, realistic measure of where current models stand relative to the agentic capabilities businesses actually need.

Daniel Shepard, Robin Salimans

LLM Agent EvaluationBusiness Process AutomationAI Agent BenchmarksWorkflow AutomationTool UseAutomated Agent Evaluation
AskPDF
Apr 20, 2026

AI scientists produce results without reasoning scientifically

Large language model (LLM)-based systems are increasingly deployed to conduct scientific research autonomously, yet whether their reasoning adheres to the epistemic norms that make scientific inquiry self-correcting is poorly understood. Here, we evaluate LLM-based scientific agents across eight domains, spanning workflow execution to hypothesis-driven inquiry, through more than 25,000 agent runs and two complementary lenses: (i) a systematic performance analysis that decomposes the contributions of the base model and the agent scaffold, and (ii) a behavioral analysis of the epistemological structure of agent reasoning. We observe that the base model is the primary determinant of both performance and behavior, accounting for 41.4% of explained variance versus 1.5% for the scaffold. Across all configurations, evidence is ignored in 68% of traces, refutation-driven belief revision occurs in 26%, and convergent multi-test evidence is rare. The same reasoning pattern appears whether the agent executes a computational workflow or conducts hypothesis-driven inquiry. They persist even when agents receive near-complete successful reasoning trajectories as context, and the resulting unreliability compounds across repeated trials in epistemically demanding domains. Thus, current LLM-based agents execute scientific workflows but do not exhibit the epistemic patterns that characterize scientific reasoning. Outcome-based evaluation cannot detect these failures, and scaffold engineering alone cannot repair them. Until reasoning itself becomes a training target, the scientific knowledge produced by such agents cannot be justified by the process that generated it.

Martiño Ríos-García, Nawaf Alampara, Chandan Gupta +5

Laboratory of Organic and Macromolecular Chemistry (IOMC), Friedrich Schiller University Jena, Humboldstr. 10, 07743 Jena, Germany · Department of Civil Engineering, Indian Institute of Technology Delhi, Hauz Khas, New Delhi 110016, India · School of Interdisciplinary Research, Indian Institute of Technology Delhi, Hauz Khas, New Delhi 110016, India +3

AI Agent ReliabilityScientific ReasoningLLM Agent EvaluationLLM ReliabilityAutonomous Scientific DiscoveryScientific Reasoning in Language Models+2
AskPDF
Apr 20, 2026

AJ-Bench: Benchmarking Agent-as-a-Judge for Environment-Aware Evaluation

As reinforcement learning continues to scale the training of large language model-based agents, reliably verifying agent behaviors in complex environments has become increasingly challenging. Existing approaches rely on rule-based verifiers or LLM-as-a-Judge models, which struggle to generalize beyond narrow domains. Agent-as-a-Judge addresses this limitation by actively interacting with environments and tools to acquire verifiable evidence, yet its capabilities remain underexplored. We introduce a benchmark AJ-Bench to systematically evaluate Agent-as-a-Judge across three domains-search, data systems, and graphical user interfaces-comprising 155 tasks and 516 annotated trajectories. The benchmark comprehensively assesses judge agents' abilities in information acquisition, state verification, and process verification. Experiments demonstrate consistent performance gains over LLM-as-a-Judge baselines, while also revealing substantial open challenges in agent-based verification. Our data and code are available at https://aj-bench.github.io/.

Wentao Shi, Yu Wang, Yuyang Zhao +8

University of Science and Technology of China · National University of Singapore · Meituan

LLM-as-a-JudgeLLM Agent EvaluationAI Agent EvaluationLLM Agent VerificationAI Agent Benchmarks
AskPDF
Apr 20, 2026

WebCompass: Towards Multimodal Web Coding Evaluation for Code Language Models

Large language models are rapidly evolving into interactive coding agents capable of end-to-end web coding, yet existing benchmarks evaluate only narrow slices of this capability, typically text-conditioned generation with static-correctness metrics, leaving visual fidelity, interaction quality, and codebase-level reasoning largely unmeasured. We introduce WebCompass, a multimodal benchmark that provides unified lifecycle evaluation of web engineering capability. Recognizing that real-world web coding is an iterative cycle of generation, editing, and repair, WebCompass spans three input modalities (text, image, video) and three task types (generation, editing, repair), yielding seven task categories that mirror professional workflows. Through a multi-stage, human-in-the-loop pipeline, we curate instances covering 15 generation domains, 16 editing operation types, and 11 repair defect types, each annotated at Easy/Medium/Hard levels. For evaluation, we adopt a checklist-guided LLM-as-a-Judge protocol for editing and repair, and propose a novel Agent-as-a-Judge paradigm for generation that autonomously executes generated websites in a real browser, explores interactive behaviors via the Model Context Protocol (MCP), and iteratively synthesizes targeted test cases, closely approximating human acceptance testing. We evaluate representative closed-source and open-source models and observe that: (1) closed-source models remain substantially stronger and more balanced; (2) editing and repair exhibit distinct difficulty profiles, with repair preserving interactivity better but remaining execution-challenging; (3) aesthetics is the most persistent bottleneck, especially for open-source models; and (4) framework choice materially affects outcomes, with Vue consistently challenging while React and Vanilla/HTML perform more strongly depending on task type.

Xinping Lei, Xinyu Che, Junqi Xiong +16

Nanjing University · Kuaishou Technology

LLM Agent EvaluationCode Language ModelsCode Generation Benchmarks
AskPDF
Apr 20, 2026

Do LLMs Need to See Everything? A Benchmark and Study of Failures in LLM-driven Smartphone Automation using Screentext vs. Screenshots

With the rapid advancement of large language models (LLMs), mobile agents have emerged as promising tools for phone automation, simulating human interactions on screens to accomplish complex tasks. However, these agents often suffer from low accuracy, misinterpretation of user instructions, and failure on challenging tasks, with limited prior work examining why and where they fail. To address this, we introduce DailyDroid, a benchmark of 75 tasks in five scenarios across 25 Android apps, spanning three difficulty levels to mimic everyday smartphone use. We evaluate it using text-only and multimodal (text + screenshot) inputs on GPT-4o and o4-mini across 300 trials, revealing comparable performance with multimodal inputs yielding marginally higher success rates. Through in-depth failure analysis, we compile a handbook of common failures. Our findings reveal critical issues in UI accessibility, input modalities, and LLM/app design, offering implications for future mobile agents, applications, and UI development.

Shiquan Zhang, Tianyi Zhang, Le Fang +3

School of Computing and Information Systems, University of Melbourne, Melbourne, Australia · School of Computer Science, University of Auckland, Auckland, New Zealand

Mobile GUI AutomationLLM Agent EvaluationAI Agent BenchmarksMultimodal AgentsLLM Agent Reliability
AskPDF
Apr 19, 2026

Agents Explore but Agents Ignore: LLMs Lack Environmental Curiosity

LLM-based agents are assumed to integrate environmental observations into their reasoning: discovering highly relevant but unexpected information should naturally lead to a model exploiting its own discoveries. We show that this assumption is false for current LLM-based agents, which struggle to reflect or react to unexpected information. Across three benchmarks (Terminal-Bench, SWE-Bench, AppWorld), we inject complete task solutions into the agent environments to deliberately expose a task's solution to a model. While agents discover these solutions on Terminal-Bench in 79-81% of runs, they interact, or exploit, them in only 37-50% of cases. This gap is starkest in AppWorld: agents see documentation stating that a command "returns the complete solution to this task" in over 90% of attempts but exploit this in fewer than 7% of trials. We show that agents lack what we call environmental curiosity: the capability to recognize and investigate unexpected but relevant observations in response to environmental stimuli. We identify three main factors influencing environmental curiosity: available tools in the agent scaffold, test-time compute, and training data distribution. Our findings identify configurations that maximize curiosity also achieve the best performance on the unmodified benchmarks. Yet even jointly optimized agents still ignore discovered solutions in the majority of trials: current agents use the environment to fetch expected information, but not to revise their strategy or maximally exploit useful stimuli.

Leon Engländer, Sophia Althammer, Ahmet Üstün +2

Cohere · Poolside

LLM Agent EvaluationTool-Using AgentsLLM Agent Reliability
AskPDF
Apr 19, 2026

HalluClear: Diagnosing, Evaluating and Mitigating Hallucinations in GUI Agents

While progress in GUI agents has been largely driven by industrial-scale training, ungrounded hallucinations often trigger cascading failures in real-world deployments.Unlike general VLM domains, the GUI agent field lacks a hallucination-focused suite for fine-grained diagnosis, reliable evaluation, and targeted mitigation.To bridge this gap, we introduce HalluClear, a comprehensive suite for hallucination mitigation in GUI agents as a complement to computation-intensive scaling. HalluClear comprises: (1) a GUI-specific hallucination taxonomy derived from empirical failure analysis; (2) a calibrated three-stage evaluation workflow which enhances VLM-as-a-judge reliability via expert-annotated benchmarking and ensemble credibility estimation; and (3) a mitigation scheme based on closed-loop structured reasoning, enabling lightweight continual post-training with cold-start initialization for both generalist and GUI-specialist agents. Experiments across representative agents and public benchmarks demonstrate that post-training on only 9K samples within our suite can significantly reduce hallucinations, thereby improving grounding and action fidelity, offering a compute-efficient pathway to robust GUI automation.

Chao Jin, Wenkui Yang, Hao Sun +6

MAIS&NLPR, Institute of Automation, Chinese Academy of Sciences

GUI AgentsHallucination in Language ModelsLLM Agent EvaluationAI Agent BenchmarksHallucination Mitigation
AskPDF
Apr 18, 2026

Systematic Capability Benchmarking of Frontier Large Language Models for Offensive Cyber Tasks

We present, to our knowledge, the most comprehensive cross-model evaluation of LLM agents on offensive cybersecurity tasks, benchmarking 10 frontier models from 7 providers on all 200 challenges of the NYU CTF Bench. Building on the D-CIPHER multi-agent framework, we extend it with multi-provider backend support, a custom Kali Linux environment with over 100 pre-installed penetration testing tools, and runtime tool-discovery agents. Through a controlled factorial study, we find that the Kali Linux environment yields a +9.5 percentage-point improvement over Ubuntu, while auto-prompting and category-specific tips often degrade performance in well-equipped environments. Among models, Claude 4.5 Opus achieves the highest solve rate (59%), followed by Gemini 3 Pro (52%), with Gemini 3 Flash offering the best cost-efficiency at $0.05 per solve. Asymmetric planner/executor model assignments provide no meaningful benefit while coherent same-model configurations consistently outperform mixed-tier pairings. Our results indicate that environment tooling and model selection emerge as the strongest drivers of performance, whereas prompt engineering interventions show diminishing or negative returns in well-equipped environments. Reported performance reflects both model reasoning ability and compatibility with agent tooling and API integration.

Tyler H. Merves, Michael H. Conaway, Joseph M. Escobar +2

Department of Cybersecurity · Department of Information Sciences and Technology

LLM Agent EvaluationCybersecurityAI Agent Security BenchmarksLLMs for Cybersecurity
AskPDF
Apr 18, 2026

If Only My CGM Could Speak: A Privacy-Preserving Agent for Question Answering over Continuous Glucose Data

Continuous glucose monitors (CGMs) used in diabetes care collect rich personal health data that could improve day-to-day self-management. However, current patient platforms only offer static summaries which do not support inquisitive user queries. Large language models (LLMs) could enable free-form inquiries about continuous glucose data, but deploying them over sensitive health records raises privacy and accuracy concerns. In this paper, we present CGM-Agent, a privacy-preserving framework for question answering over personal glucose data. In our design, the LLM serves purely as a reasoning engine that selects analytical functions. All computation occurs locally, and personal health data never leaves the user's device. For evaluation, we construct a benchmark of 4,180 questions combining parameterized question templates with real user queries and ground truth derived from deterministic program execution. Evaluating 6 leading LLMs, we find that top models achieve 94% value accuracy on synthetic queries and 88% on ambiguous real-world queries. Errors stem primarily from intent and temporal ambiguity rather than computational failures. Additionally, lightweight models achieve competitive performance in our agent design, suggesting opportunities for low-cost deployment. We release our code and benchmark to support future work on trustworthy health agents.

Yanjun Cui, Ali Emami, Temiloluwa Prioleau +1

Dartmouth College · Emory University

Privacy-Preserving Language ModelsHealthcareLLM Agent EvaluationBiomedical QAWearable Health MonitoringContinuous Glucose Monitoring
AskPDF
Apr 17, 2026

Don't Start What You Can't Finish: A Counterfactual Audit of Support-State Triage in LLM Agents

Current agent evaluations largely reward execution on fully specified tasks, while recent work studies clarification [11, 22, 2], capability awareness [9, 1], abstention [8, 14], and search termination [20, 5] mostly in isolation. This leaves open whether agents can diagnose why a task is blocked before acting. We introduce the Support-State Triage Audit (SSTA-32), a matched-item diagnostic framework in which minimal counterfactual edits flip the same base request across four support states: Complete (ANSWER), Clarifiable (CLARIFY), Support-Blocked (REQUEST SUPPORT), and Unsupported-Now (ABSTAIN). We evaluate a frontier model under four prompting conditions - Direct, Action-Only, Confidence-Only, and a typed Preflight Support Check (PSC) - using Dual-Persona Auto-Auditing (DPAA) with deterministic heuristic scoring. Default execution overcommits heavily on non-complete tasks (41.7% overcommitment rate). Scalar confidence mapping avoids overcommitment but collapses the three-way deferral space (58.3% typed deferral accuracy). Conversely, both Action-Only and PSC achieve 91.7% typed deferral accuracy by surfacing the categorical ontology in the prompt. Targeted ablations confirm that removing the support-sufficiency dimension selectively degrades REQUEST SUPPORT accuracy, while removing the evidence-sufficiency dimension triggers systematic overcommitment on unsupported items. Because DPAA operates within a single context window, these results represent upper-bound capability estimates; nonetheless, the structural findings indicate that frontier models possess strong latent triage capabilities that require explicit categorical decision paths to activate safely.

Eren Unlu

Globeholder Paris, France

LLM Agent EvaluationLLM Refusal BehaviorLLM Decision-MakingLLM Agent Reliability
AskPDF
Apr 17, 2026

Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench

Automated evaluation of tool-using large language model (LLM) agents is widely assumed to be reliable, but this assumption has rarely been validated against human annotation. We introduce AgentProp-Bench, a 2,000-task benchmark with 2,300 traces across four domains, nine production LLMs, and a 100-label human-validated subset. We quantify judge reliability, characterize error propagation, and evaluate a runtime mitigation. Substring-based judging agrees with human annotation at kappa=0.049 (chance-level); a three-LLM ensemble reaches kappa=0.432 (moderate) with a conservative bias. Under validated evaluation, a parameter-level injection propagates to a wrong final answer with human-calibrated probability approximately 0.62 (range 0.46-0.73 across models). Rejection (catching bad parameters) and recovery (correcting after acceptance) are independent model capabilities (Spearman rho=0.126, p=0.747). A tuned runtime interceptor reduces hallucination on GPT-4o-mini by 23.0 percentage points under a concurrent n=600 control, but shows no significant effect on Gemini-2.0-Flash, whose aggressive parameter rejection eliminates the target failure mode. All code, data, traces, and human labels are released at https://github.com/bhaskargurram-ai/agenthallu-bench.

Bhaskar Gurram

Zasti Inc., Ashburn, VA, USA

AI Agent ReliabilityTool-Augmented Language Model AgentsLLM-as-a-JudgeLLM Agent EvaluationAI Agent BenchmarksLLM Reliability+2
AskPDF
Apr 17, 2026

SocialGrid: A Benchmark for Planning and Social Reasoning in Embodied Multi-Agent Systems

As Large Language Models (LLMs) transition from text processors to autonomous agents, evaluating their social reasoning in embodied multi-agent settings becomes critical. We introduce SocialGrid, an embodied multi-agent environment inspired by Among Us that evaluates LLM agents on planning, task execution, and social reasoning. Our evaluations reveal that even the strongest open model (GPT-OSS-120B) achieves below 60% accuracy in task completion and planning, with agents getting stuck in repetitive behaviors or failing to navigate basic obstacles. Since poor navigation confounds evaluation of social intelligence, SocialGrid offers an optional Planning Oracle to isolate social reasoning from planning deficits. While planning assistance improves task completion, social reasoning remains a bottleneck: agents fail to detect deception at near-random chance regardless of scale, relying on shallow heuristics rather than accumulating behavioral evidence. SocialGrid provides automatic failure analysis and fine-grained metrics, enabling developers to diagnose and improve their agents. We also establish a competitive leaderboard using Elo ratings from adversarial league play.

Hikaru Shindo, Hanzhao Lin, Lukas Helff +2

Technical University of Darmstadt · German Research Center for Artificial Intelligence · CERTAIN +2

LLM Agent EvaluationAI Agent BenchmarksMulti-Agent SystemsMulti-Agent ReasoningLLM Agent Planning
AskPDF
Apr 16, 2026

Understanding Tool-Augmented Agents for Lean Formalization: A Factorial Analysis

Automatic translation of natural language mathematics into faithful Lean 4 code is hindered by the fundamental dissonance between informal set-theoretic intuition and strict formal type theory. This gap often causes LLMs to hallucinate non-existent library definitions, resulting in code that fails to compile or lacks semantic fidelity. In this work, we investigate the effectiveness of tool-augmented agents for this task through a systematic factorial analysis of three distinct tool categories: Fine-tuned Model Querying (accessing expert drafts), Knowledge Search (retrieving symbol definitions), and Compiler Feedback (verifying code via a Lean REPL). We first benchmark the agent against one-shot baselines, demonstrating large gains in both compilation success and semantic equivalence. We then use the factorial decomposition to quantify the impact of each category, isolating the marginal contribution of each tool type to overall performance.

Ke Zhang, Patricio Gallardo, Maziar Raissi +1

University of California, Riverside.

Code TranslationLLM Agent EvaluationLean Theorem ProvingAutoformalizationTool-Using Language Agents
AskPDF
Apr 16, 2026

CoopEval: Benchmarking Cooperation-Sustaining Mechanisms and LLM Agents in Social Dilemmas

It is increasingly important that LLM agents interact effectively and safely with other goal-pursuing agents, yet, recent works report the opposite trend: LLMs with stronger reasoning capabilities behave less cooperatively in mixed-motive games such as the prisoner's dilemma and public goods settings. Indeed, our experiments show that recent models -- with or without reasoning enabled -- consistently defect in single-shot social dilemmas. To tackle this safety concern, we present the first comparative study of game-theoretic mechanisms designed to enable cooperative outcomes between rational agents in equilibrium. Across four social dilemmas testing distinct components of robust cooperation, we evaluate four families of mechanisms: (1) repeating the game for many rounds, (2) reputation systems, (3) third-party mediators to delegate decision making to, and (4) contract agreements for outcome-conditional payments between players. Among our findings, we establish that contracting and mediation are most effective in achieving cooperative outcomes between capable LLM models, and that repetition-induced cooperation deteriorates drastically when co-players vary. Moreover, we demonstrate that the mechanisms become more effective under evolutionary pressures to maximize individual payoffs.

Emanuel Tewolde, Xiao Zhang, David Guzman Piedrahita +2

Carnegie Mellon University · Foundations of Cooperative AI Lab (FOCAL) · Jinesis Lab, University of Toronto & Vector Institute +3

Social DilemmasCooperative Game TheoryMulti-Agent CoordinationMechanism DesignGame TheoryLLM Agent Evaluation+1
AskPDF
Apr 16, 2026

From Reactive to Proactive: Assessing the Proactivity of Voice Agents via ProVoice-Bench

Recent advancements in LLM agents are gradually shifting from reactive, text-based paradigms toward proactive, multimodal interaction. However, existing benchmarks primarily focus on reactive responses, overlooking the complexities of proactive intervention and monitoring. To bridge this gap, we introduce ProVoice-Bench, the first evaluation framework specifically designed for proactive voice agents, featuring four novel tasks. By leveraging a multi-stage data synthesis pipeline, we curate 1,182 high-quality samples for rigorous testing. Our evaluation of state-of-the-art Multimodal LLMs reveals a significant performance gap, particularly regarding over-triggering and reasoning capabilities. These findings highlight the limitations of current models and offer a roadmap for developing more natural, context-aware proactive agents.

Ke Xu, Yuhao Wang, Yu Wang

Shanghai Jiao Tong University

LLM Agent EvaluationAgent EvaluationAI Agent BenchmarksMultimodal Model Evaluation
AskPDF
Apr 16, 2026

Task-Level AI Readiness Assessment for Business Process Management:The T-IPO Model and LARA Matrix in Financial-Services IT Operations

Which tasks inside an enterprise workflow can a large-language-model agent reliably handle, and under what conditions? Most business process modeling frameworks still answer this at the activity level, even though a single activity can bundle work of radically different difficulty. This paper takes the analysis a step smaller. We describe two design artifacts developed in a financial-services IT setting: T-IPO, which represents each task as an eight-element tuple, and LARA (LLM Agent Readiness Assessment), a five-dimension rubric that scores a task's readiness for agent substitution. Compliance Sensitivity carries 1.5×1.5\times weight, a value we fixed through a three-round Delphi study and cross-checked with AHP. The rubric produces four levels, L1 to L4, and applies a floor rule so that a task with maximum compliance load cannot be classified below L3 no matter what the other scores say. Both artifacts sit inside a larger methodology (PARTIS) that we map onto BWW ontology in Section 3. We evaluate the instruments across 127 tasks. Inter-rater agreement reaches Fleiss' κ=0.80κ= 0.80; a replication at three further institutions returns κ=0.73κ= 0.73. A controlled comparison against activity-level assessment suggests, though does not prove, an improvement in predictive utility at the task level. Pilot deployment of 120 task instances confirms that auto-completion decays monotonically from 95%95\% at L1 through about 70%70\% at L2 to about 40%40\% at L3. Exploratory factor analysis points to a two-factor structure: task readiness seems to be determined jointly by cognitive-execution complexity and governance-compliance intensity. We close with a recalibration procedure (LARA-TCA) so the rubric can keep pace with evolving LLM capabilities.

Mingjun Li, Xiaojun Ye

Tsinghua University, Beijing, China

LLM Agent EvaluationMulti-Criteria Decision AnalysisBusiness Process Modeling
AskPDF
Apr 16, 2026

HWE-Bench: Benchmarking LLM Agents on Real-World Hardware Bug Repair Tasks

Existing benchmarks for hardware design primarily evaluate Large Language Models (LLMs) on isolated, component-level tasks such as generating HDL modules from specifications, leaving repository-scale evaluation unaddressed. We introduce HWE-Bench, the first large-scale, repository-level benchmark for evaluating LLM agents on real-world hardware bug repair tasks. HWE-Bench comprises 417 task instances derived from real historical bug-fix pull requests across six major open-source projects spanning both Verilog/SystemVerilog and Chisel, covering RISC-V cores, SoCs, and security roots-of-trust. Each task is grounded in a fully containerized environment where the agent must resolve a real bug report, with correctness validated through the project's native simulation and regression flows. The benchmark is built through a largely automated pipeline that enables efficient expansion to new repositories. We evaluate seven LLMs with four agent frameworks and find that the best agent resolves 70.7% of tasks overall, with performance exceeding 90% on smaller cores but dropping below 65% on complex SoC-level projects. We observe larger performance gaps across models than commonly reported on software benchmarks, and difficulty is driven by project scope and bug-type distribution rather than code size alone. Our failure analysis traces agent failures to three stages of the debugging process: fault localization, hardware-semantic reasoning, and cross-artifact coordination across RTL, configuration, and verification components, providing concrete directions for developing more capable hardware-aware agents.

Fan Cui, Hongyuan Hou, Zizhang Luo +2

Peking University · Beijing, China

Automated Program RepairElectronic Design AutomationLLM Agent EvaluationAI Agent Benchmarks
AskPDF
Apr 16, 2026

DR3^{3}-Eval: Towards Realistic and Reproducible Deep Research Evaluation

Deep Research Agents (DRAs) aim to solve complex, long-horizon research tasks involving planning, retrieval, multimodal understanding, and report generation, yet their evaluation remains challenging due to dynamic web environments and ambiguous task definitions. We propose DR3^{3}-Eval, a realistic and reproducible benchmark for evaluating deep research agents on multimodal, multi-file report generation. DR3^{3}-Eval is constructed from authentic user-provided materials and paired with a per-task static research sandbox corpus that simulates open-web complexity while remaining fully verifiable, containing supportive documents, distractors, and noise. Moreover, we introduce a multi-dimensional evaluation framework measuring Information Recall, Factual Accuracy, Citation Coverage, Instruction Following, and Depth Quality, and validate its alignment with human judgments. Experiments with our developed multi-agent system DR3^{3}-Agent based on multiple state-of-the-art language models demonstrate that DR3^{3}-Eval is highly challenging and reveals critical failure modes in retrieval robustness and hallucination control. Our code and data are publicly available.

Qianqian Xie, Qingheng Xiong, He Zhu +16

Nanjing University · M-A-P · Nanjing University of Science and Technology +2

LLM Agent EvaluationLong-Horizon Agent EvaluationAI Agent Benchmarks
AskPDF
Apr 15, 2026

GeoAgentBench: A Dynamic Execution Benchmark for Tool-Augmented Agents in Spatial Analysis

The integration of Large Language Models (LLMs) into Geographic Information Systems (GIS) marks a paradigm shift toward autonomous spatial analysis. However, evaluating these LLM-based agents remains challenging due to the complex, multi-step nature of geospatial workflows. Existing benchmarks primarily rely on static text or code matching, neglecting dynamic runtime feedback and the multimodal nature of spatial outputs. To address this gap, we introduce GeoAgentBench (GABench), a dynamic and interactive evaluation benchmark tailored for tool-augmented GIS agents. GABench provides a realistic execution sandbox integrating 117 atomic GIS tools, encompassing 53 typical spatial analysis tasks across 6 core GIS domains. Recognizing that precise parameter configuration is the primary determinant of execution success in dynamic GIS environments, we designed the Parameter Execution Accuracy (PEA) metric, which utilizes a "Last-Attempt Alignment" strategy to quantify the fidelity of implicit parameter inference. Complementing this, a Vision-Language Model (VLM) based verification is proposed to assess data-spatial accuracy and cartographic style adherence. Furthermore, to address the frequent task failures caused by parameter misalignments and runtime anomalies, we developed a novel agent architecture, Plan-and-React, that mimics expert cognitive workflows by decoupling global orchestration from step-wise reactive execution. Extensive experiments with seven representative LLMs demonstrate that the Plan-and-React paradigm significantly outperforms traditional frameworks, achieving the optimal balance between logical rigor and execution robustness, particularly in multi-step reasoning and error recovery. Our findings highlight current capability boundaries and establish a robust standard for assessing and advancing the next generation of autonomous GeoAI.

Bo Yu, Cheng Yang, Dongyang Hou +6

School of Geosciences and Info-Physics, Central South University, Changsha, China. · School of Earth Sciences and Spatial Information Engineering, Hunan University of Science and Technology, Xiangtan, China.

LLM Agent EvaluationAI Agent BenchmarksLLM Agent OrchestrationSpatial Reasoning
AskPDF
Apr 14, 2026

Exploration and Exploitation Errors Are Measurable for Language Model Agents

Language Model (LM) agents are increasingly used in complex decision-making tasks, from AI coding to physical AI, where they must both explore a search space and exploit acquired knowledge effectively. Although exploration and exploitation play different roles in decision-making, it remains challenging to systematically distinguish exploration and exploitation without access to the agent's internal policy, and to determine when either constitutes an error. To address this, we design partially observable 2D grid maps paired with an unknown symbolic task directed acyclic graph (DAG), abstracting practical agentic AI scenarios. We then propose a policy-agnostic metric that detects unproductive traversals, while permitting necessary backtracking. We attribute each error to exploration or exploitation based on whether the current state admits productive exploration, exploitation, or both. Using generated maps and task DAGs with controllable difficulty, we evaluate 35 frontier LM agents and find that even state-of-the-art models struggle to reach the goal within the step budget. log⁡\log exploration and exploitation errors exhibit strong relationships with task success (R2=0.885R^2 = 0.885 and 0.8750.875, respectively), providing empirical evidence that the metric captures behavior relevant to task success. Beyond success rate, the metric reveals distinct exploration and exploitation behaviors among models with similar success rates. We further find that a compact memory harness achieves higher task success than a reasoning baseline that retains its own reasoning traces. Finally, we extend our metric to an iconic video game, Minecraft Clone, beyond our synthetic environment. We release our code at https://github.com/jjj-madison/measurable-explore-exploit.

Jaden Park, Jungtaek Kim, Jongwon Jeong +3

University of Wisconsin–Madison · KRAFTON · Ludo Robotics

Exploration-Exploitation TradeoffLLM Agent EvaluationAI Agent Benchmarks
AskPDF
Apr 14, 2026

Grounded autonomous scrutiny at scale: emergent critique from reproduction of published computational physics papers

Autonomous LLM agents now produce complete research artifacts in machine-learning sandboxes, but real computational physics is harder: experiments are first-principles calculations against re-runnable physical ground truth, and meaningful new work almost always builds on a key existing paper. We ask whether such an agent can perform grounded scrutiny of published computational physics - reading a paper, reproducing it from scratch, and surfacing methodological concerns from execution. We deploy a single Claude Opus 4.6 configuration at two complementary scopes. At scale, across 111 open-access Quantum ESPRESSO papers, an autonomous agent runs the read-plan-compute-compare loop and, although never asked to critique, raises substantive methodological concerns on ~42% of papers; 85 of 88 of these critiques (96.6%) surface only after the agent has actually run a calculation, with a reading-only ceiling of 1.8%. Critique emerges from reproduction, not from reading. In depth, on one Nature Communications paper on multiscale device simulation of a 2D-material MOSFET, a fresh agent inheriting a verified reproduction pipeline autonomously produces a 14-concern physics inventory and a complete, submission-form six-page Comment that revises the paper's L_G = 5 nm headline. Two of its L_G = 5 nm headline-challenging attacks - a source-degeneration contact-resistance bound and a Sb-doping degradation ratio - are absent from the published 21-reviewer peer review.

Haonan Huang

Department of Physics, Princeton University

LLM Agent EvaluationScientific ReproducibilityAutomated Peer ReviewAutonomous Research Agents
AskPDF
Apr 13, 2026

From Plan to Action: How Well Do Agents Follow the Plan?

Agents are commonly instructed to follow a task-specific plan for guidance. However, it is unknown to what extent agents actually follow instructed plans. Without such an analysis, determining the extent agents comply with a given plan, it is impossible to assess whether a solution was reached through correct strategic reasoning or through other means, e.g., data contamination or overfitting to a benchmark. This paper presents the first extensive, systematic analysis of plan compliance in programming agents, examining 21,120 trajectories from SWE-agent across four LLMs on SWE-bench Verified and SWE-bench Pro under eight plan variations. Without an explicit plan, agents fall back on internalized workflows during training, which are often incomplete, overfit, or inconsistently applied. Providing the standard plan improves issue resolution, and we observe that periodic plan reminders can mitigate plan violations and improve task success. A subpar plan hurts performance even more than no plan at all. Surprisingly, inserting additional task-relevant phases in the early stage can degrade performance, particularly when these phases do not align with the model's internal problem-solving strategy. These findings call for fine-tuning paradigms that teach models to follow instructed plans, rather than encoding task-specific plans in them, so that they reason and act adaptively, rather than memorizing workflows.

Shuyang Liu, Saman Dehghan, Jatin Ganhotra +2

University of Illinois Urbana-Champaign Urbana, USA · IBM New York, USA

Software Engineering AgentsLLM Agent PlanningLLM Agent Evaluation
AskPDF
Apr 13, 2026

The A-R Behavioral Space: Execution-Level Profiling of Tool-Using Language Model Agents in Organizational Deployment

Large language models (LLMs) are increasingly deployed as tool-augmented agents capable of executing system-level operations. While existing benchmarks primarily assess textual alignment or task success, less attention has been paid to the structural relationship between linguistic signaling and executable behavior under varying autonomy scaffolds. This study introduces an execution-layer be-havioral measurement approach based on a two-dimensional A-R space defined by Action Rate (A) and Refusal Signal (R), with Divergence (D) capturing coor-dination between the two. Models are evaluated across four normative regimes (Control, Gray, Dilemma, and Malicious) and three autonomy configurations (di-rect execution, planning, and reflection). Rather than assigning aggregate safety scores, the method characterizes how execution and refusal redistribute across contextual framing and scaffold depth. Empirical results show that execution and refusal constitute separable behavioral dimensions whose joint distribution varies systematically across regimes and autonomy levels. Reflection-based scaffolding often shifts configurations toward higher refusal in risk-laden contexts, but redis-tribution patterns differ structurally across models. The A-R representation makes cross-sectional behavioral profiles, scaffold-induced transitions, and coordination variability directly observable. By foregrounding execution-layer characterization over scalar ranking, this work provides a deployment-oriented lens for analyzing and selecting tool-enabled LLM agents in organizational settings where execution privileges and risk tolerance vary.

Shasha Yu, Fiona Carroll, Barry L. Bentley

Cardiff School of Technologies, Cardiff Metropolitan University, Cardiff, Wales CF5 2YB, UK · School of Professional Studies, Clark University, Worcester, MA 01610 USA · Harvard Medical School, Harvard University, Boston, MA 02115 USA

LLM Agent EvaluationTool-Using Language AgentsLanguage Model Safety EvaluationRefusal Behavior in Language Models
AskPDF
Apr 13, 2026

RPA-Check: A Multi-Stage Automated Framework for Evaluating Dynamic LLM-based Role-Playing Agents

The rapid adoption of Large Language Models (LLMs) in interactive systems has enabled the creation of dynamic, open-ended Role-Playing Agents (RPAs). However, evaluating these agents remains a significant challenge, as standard NLP metrics fail to capture the nuances of role adherence, logical consistency, and long-term narrative stability. This paper introduces RPA-Check, a multi-stage automated evaluation framework designed to objectively assess the performance of LLM-based RPAs in complex, constraints-heavy environments. Our methodology is based on a four-step pipeline: (1) Dimension Definition, establishing high-level qualitative behavioral criteria; (2) Augmentation, where these requirements are expanded into granular boolean checklist indicators; (3) Semantic Filtering, to ensure indicator objectivity, no redundancy and agent isolation; and (4) LLM-as-a-Judge Evaluation, which employs chain-of-thought verification to score agent fidelity. We validate this framework by applying it to LLM Court, a serious game for forensic training involving several quantized local models. Experimental results across five distinct legal scenarios demonstrate the framework's ability to identify subtle trade-offs between model size, reasoning depth, and operational stability. Notably, the findings reveal an inverse relationship between parametric scale and procedural consistency, showing that smaller, adequately instruction-tuned models (8-9B) can outperform larger architectures prone to user-alignment bias or sycophancy. RPA-Check thus provides a standardized and reproducible metric for future research in generative agent evaluation within specialized domains.

Riccardo Rosati, Edoardo Colucci, Massimiliano Bolognini +2

Department of Political Sciences, Communication and International Relations, University of Macerata, Via Don Minzoni 22/A, Macerata, 62100, Italy · Deparment of Information Engineering, Università Politecnica delle Marche, Via Brecce Bianche 12, Ancona, 60131, Italy · Department of Law, University of Macerata, Piaggia dell’Università 2, Macerata, 62100, Italy

Role-Playing Agent EvaluationLLM Agent EvaluationAgent EvaluationLarge Language Model-Based Role-Play SimulationRubric-Based Evaluation
AskPDF