AI Agent Evaluation

Follow this topic Ask about this topic

AI agent evaluation studies how well systems use tools and interact with changing environments to complete tasks, from coding and web navigation to scientific and professional workflows. Researchers build interactive benchmarks and assess outcomes alongside execution traces, using deterministic state checks, expert rubrics, and language-model judges. Repeated trials and process-level measures help reveal failures in tool use, safety, consistency, and recovery that a final pass rate can hide. A central challenge is ensuring that benchmark scores reflect real capability rather than leakage, reward hacking, evaluator bias, or quirks of a particular model harness.

Latest papers

525 papers77 papers in the last four weeks, up 4% on the four weeks before. 0.4% of all new papers.

Mar 10, 2026

Video-Based Reward Modeling for Computer-Use Agents

Computer-using agents (CUAs) are becoming increasingly capable; however, it remains difficult to scale evaluation of whether a trajectory truly fulfills a user instruction. In this work, we study reward modeling from execution video: a sequence of keyframes from an agent trajectory that is independent of the agent's internal reasoning or actions. Although video-execution modeling is method-agnostic, it presents key challenges, including highly redundant layouts and subtle, localized cues that determine success. We introduce Execution Video Reward 53k (ExeVR-53k), a dataset of 53k high-quality video--task--reward triplets. We further propose adversarial instruction translation to synthesize negative samples with step-level annotations. To enable learning from long, high-resolution execution videos, we design spatiotemporal token pruning, which removes homogeneous regions and persistent tokens while preserving decisive UI changes. Building on these components, we fine-tune an Execution Video Reward Model (ExeVRM) that takes only a user instruction and a video-execution sequence to predict task success. Our ExeVRM 8B achieves 84.7% accuracy and 87.7% recall on video-execution assessment, outperforming strong proprietary models such as GPT-5.2 and Gemini-3 Pro across Ubuntu, macOS, Windows, and Android, while providing more precise temporal attribution. These results show that video-execution reward modeling can serve as a scalable, model-agnostic evaluator for CUAs.

Linxin Song, Jieyu Zhang, Huanxin Sheng +6

University of Southern California · University of Washington · MBZUAI +1

Computer-Use Agent Evaluation Reward Modeling AI Agent Evaluation Multimodal Reward Modeling Computer-Use Agents

Mar 10, 2026

Beyond Scaling: Assessing Strategic Reasoning and Rapid Decision-Making Capability of LLMs in Zero-sum Environments

Large Language Models (LLMs) have achieved strong performance on static reasoning benchmarks, yet their effectiveness as interactive agents operating in adversarial, time-sensitive environments remains poorly understood. Existing evaluations largely treat reasoning as a single-shot capability, overlooking the challenges of opponent-aware decision-making, temporal constraints, and execution under pressure. This paper introduces Strategic Tactical Agent Reasoning (STAR) Benchmark, a multi-agent evaluation framework that assesses LLMs through 1v1 zero-sum competitive interactions, framing reasoning as an iterative, adaptive decision-making process. STAR supports both turn-based and real-time settings, enabling controlled analysis of long-horizon strategic planning and fast-paced tactical execution within a unified environment. Built on a modular architecture with a standardized API and fully implemented execution engine, STAR facilitates reproducible evaluation and flexible task customization. To move beyond binary win-loss outcomes, we introduce a Strategic Evaluation Suite that assesses not only competitive success but also the quality of strategic behavior, such as execution efficiency and outcome stability. Extensive pairwise evaluations reveal a pronounced strategy-execution gap: while reasoning-intensive models dominate turn-based settings, their inference latency often leads to inferior performance in real-time scenarios, where faster instruction-tuned models prevail. These results show that strategic intelligence in interactive environments depends not only on reasoning depth, but also on the ability to translate plans into timely actions, positioning STAR as a principled benchmark for studying this trade-off in competitive, dynamic settings.

Yang Li, Xing Chen, Yutao Liu +5

Tsinghua University · Ocean University of China (OUC) · CAICT +4

AI Agent Evaluation AI Agent Benchmarks Strategic Reasoning Risks in Language Models Long-Horizon Agent Evaluation

Mar 9, 2026

MASEval: Extending Multi-Agent Evaluation from Models to Systems

The rapid adoption of LLM-based agentic systems has produced a rich ecosystem of frameworks (smolagents, LangGraph, AutoGen, CAMEL, LlamaIndex, i.a.). Yet existing benchmarks are model-centric: they fix the agentic setup and do not compare other system components. We argue that implementation decisions substantially impact performance, including choices such as topology, orchestration logic, and error handling. MASEval addresses this evaluation gap with a framework-agnostic library that treats the entire system as the unit of analysis. Through a systematic system-level comparison across 3 benchmarks, 3 models, and 3 frameworks, we find that framework choice matters as much as model choice. MASEval allows researchers to explore all components of agentic systems, opening new avenues for principled system design, and practitioners to identify the best implementation for their use case. MASEval is available under the MIT licence https://github.com/parameterlab/MASEval.

Cornelius Emde, Alexander Rubinstein, Anmol Goel +4

Parameter Lab · University of Oxford · University of Tübingen +4

AI Agent Evaluation AI Agent Benchmarks LLM Agent Orchestration

Mar 9, 2026

Test-Driven AI Agent Definition (TDAD): Compiling Tool-Using Agents from Behavioral Specifications

We present Test-Driven AI Agent Definition (TDAD), a methodology that treats agent prompts as compiled artifacts: engineers provide behavioral specifications, a coding agent converts them into executable tests, and a second coding agent iteratively refines the prompt until tests pass. Deploying tool-using LLM agents in production requires measurable behavioral compliance that current development practices cannot provide. Small prompt changes cause silent regressions, tool misuse goes undetected, and policy violations emerge only after deployment. To mitigate specification gaming, TDAD introduces three mechanisms: (1) visible/hidden test splits that withhold evaluation tests during compilation, (2) semantic mutation testing via a post-compilation agent that generates plausible faulty prompt variants, with the harness measuring whether the test suite detects them, and (3) spec evolution scenarios that quantify regression safety when requirements change. We evaluate TDAD on SpecSuite-Core, a benchmark of four deeply-specified agents spanning policy compliance, grounded analytics, runbook adherence, and deterministic enforcement. Across 24 independent trials, TDAD achieves 92% v1 compilation success with 97% mean hidden pass rate; evolved specifications compile at 58%, with most failed runs passing all visible tests except 1-2, and show 86-100% mutation scores, 78% v2 hidden pass rate, and 97% regression safety scores. The implementation is available as an open benchmark at https://github.com/f-labs-io/tdad-paper-code.

Tzafrir Rehan

Fiverr Labs

Tool-Using Agents AI Agent Evaluation Specification-Based Testing AI Agent Benchmarks

Mar 9, 2026

ConflictBench: Evaluating Human-AI Conflict via Interactive and Visually Grounded Environments

As large language models (LLMs) evolve into autonomous agents capable of acting in open-ended environments, ensuring behavioral alignment with human values becomes a critical safety concern. Existing benchmarks, focused on static, single-turn prompts, fail to capture the interactive and multi-modal nature of real-world conflicts. We introduce ConflictBench, a benchmark for evaluating human-AI conflict through 150 multi-turn scenarios derived from prior alignment queries. ConflictBench integrates a text-based simulation engine with a visually grounded world model, enabling agents to perceive, plan, and act under dynamic conditions. Empirical results show that while agents often act safely when human harm is immediate, they frequently prioritize self-preservation or adopt deceptive strategies in delayed or low-risk settings. A regret test further reveals that aligned decisions are often reversed under escalating pressure, especially with visual input. These findings underscore the need for interaction-level, multi-modal evaluation to surface alignment failures that remain hidden in conventional benchmarks.

Weixiang Zhao, Haozhen Li, Yanyan Zhao +5

Harbin Institute of Technology · SERES

Multimodal Agent Evaluation AI Agent Safety Benchmarks AI Alignment AI Agent Evaluation Agent Evaluation

Mar 6, 2026

Evaluating Generalization Mechanisms in Autonomous Cyber Attack Agents

Autonomous offensive agents often fail to transfer beyond the networks on which they are trained. We isolate a minimal but fundamental shift -- unseen host/subnet IP reassignment in an otherwise fixed enterprise scenario -- and evaluate attacker generalization in the NetSecGame environment. Agents are trained on five IP-range variants and tested on a sixth unseen variant; only the meta-learning agent may adapt at test time. We compare three agent families (traditional RL, adaptation agents, and LLM-based agents) and use action-distribution-based behavioral/XAI analyses to localize failure modes. Some adaptation methods show partial transfer but significant degradation under unseen reassignment, indicating that even address-space changes can break long-horizon attack policies. Under our evaluation protocol and agent-specific assumptions, prompt-driven pretrained LLM agents achieve the highest success on the held-out reassignment, but at the cost of increased inference-time compute, reduced transparency, and practical failure modes such as repetition/invalid-action loops.

Ondřej Lukáš, Jihoon Shin, Emilia Rivas +6

Czech Technical University in Prague (CTU) · University of Texas at El Paso (UTEP) · Universidad Nacional de Cuyo (UNCUYO)

Generalization in RL AI Agent Evaluation LLMs for Cybersecurity

Mar 6, 2026

Evaluating Multi-Agent LLM Architectures for Rare Disease Diagnosis

While large language models are capable diagnostic tools, the impact of multi-agent topology on diagnostic accuracy remains underexplored. This study evaluates four agent topologies, Control (single agent), Hierarchical, Adversarial, and Collaborative, across 302 cases spanning 33 rare disease categories. We introduce a Reasoning Gap metric to quantify the difference between internal knowledge retrieval and final diagnostic accuracy. Results indicate that the Hierarchical topology (50.0% accuracy) marginally outperforms Collaborative (49.8%) and Control (48.5%) configurations. In contrast, the Adversarial model significantly degrades performance (27.3%), exhibiting a massive Reasoning Gap where valid diagnoses were rejected due to artificial doubt. Across all architectures, performance was strongest in Allergic diseases and Toxic Effects categories but poorest in Cardiac Malformation and Respiratory cases. Critically, while the single-agent baseline was generally robust, all multi-agent systems, including the Adversarial model, yielded superior accuracy in Bone and Thoracic disease categories. These findings demonstrate that increasing system complexity does not guarantee better reasoning, supporting a shift toward dynamic topology selection.

Ahmed Almasoud

Artificial Intelligence and Data Analytics (AIDA) Lab, College of Computing and Information Sciences, Prince Sultan University, Riyadh, Saudi Arabia.

Multi-Agent LLM Systems AI Agent Evaluation Rare Disease Diagnosis

Mar 5, 2026

TML-Bench: Benchmark for Data Science Agents on Tabular ML Tasks

Autonomous coding agents can produce strong tabular baselines quickly on Kaggle-style tasks. Practical value depends on end-to-end correctness and reliability under time limits. This paper introduces TML-Bench, a tabular benchmark for data science agents on Kaggle-style tasks. This paper evaluates 10 OSS LLMs on four Kaggle competitions and three time budgets (240s, 600s, and 1200s). Each model is run five times per task and budget. A run is successful if it produces a valid submission and a private-holdout score on hidden labels that are not accessible to the agent. This paper reports median performance, success rates, and run-to-run variability. MiniMax-M2.1 model achieves the best aggregate performance score on all four competitions under the paper's primary aggregation. Average performance improves with larger time budgets. Scaling is noisy for some individual models at the current run count. Code and materials are available at https://github.com/MykolaPinchuk/TML-bench/tree/master.

Mykola Pinchuk

Independent Researcher, San Jose, USA

AI Agent Benchmarks Agent Benchmarks AI Agent Evaluation AI Agent Reliability Tabular ML

Mar 5, 2026

AgentSCOPE: Evaluating Contextual Privacy Across Agentic Workflows

Agentic systems are increasingly acting on users' behalf, accessing calendars, email, and personal files to complete everyday tasks. Privacy evaluation for these systems has focused on the input and output boundaries, but each task involves several intermediate information flows, from agent queries to tool responses, that are not currently evaluated. We argue that every boundary in an agentic pipeline is a site of potential privacy violation and must be assessed independently. To support this, we introduce the Privacy Flow Graph, a Contextual Integrity-grounded framework that decomposes agentic execution into a sequence of information flows, each annotated with the five CI parameters, and traces violations to their point of origin. We present AgentSCOPE, a benchmark of 62 multi-tool scenarios across eight regulatory domains with ground truth at every pipeline stage. Our evaluation across seven state-of-the-art LLMs show that privacy violations in the pipeline occur in over 80% of scenarios, even when final outputs appear clean (24%), with most violations arising at the tool-response stage where APIs return sensitive data indiscriminately. These results indicate that output-level evaluation alone substantially underestimates the privacy risk of agentic systems.

Ivoline C. Ngong, Keerthiram Murugesan, Swanand Kadhe +3

University of Vermont · IBM Research

AI Agent Evaluation Agentic Workflows Contextual Integrity

Mar 4, 2026

Self-Attribution Bias: When AI Monitors Go Easy on Themselves

Agentic systems increasingly rely on language models to monitor their own behavior. For example, coding agents may self critique generated code for pull request approval or assess the safety of tool-use actions. We show that this design pattern can fail when the action is presented in a previous or in the same assistant turn instead of being presented by the user in a user turn. We define self-attribution bias as the tendency of a model to evaluate an action as more correct or less risky when the action is implicitly framed as its own, compared to when the same action is evaluated under off-policy attribution. Across four coding and tool-use datasets, we find that monitors fail to report high-risk or low-correctness actions more often when evaluation follows a previous assistant turn in which the action was generated, compared to when the same action is evaluated in a new context presented in a user turn. In contrast, explicitly stating that the action comes from the monitor does not by itself induce self-attribution bias. Because monitors are often evaluated on fixed examples rather than on their own generated actions, these evaluations can make monitors appear more reliable than they actually are in deployment, leading developers to unknowingly deploy inadequate monitors in agentic systems.

Dipika Khullar, Jack Hopkins, Rowan Wang +1

MATS · Anthropic

AI Agent Monitoring Language Model Self-Assessment AI Agent Evaluation AI Agent Safety Causal Attribution

Mar 4, 2026

A Multi-Agent Framework for Interpreting Multivariate Physiological Time Series

Continuous physiological monitoring is central to emergency care, yet deploying trustworthy AI is challenging. While LLMs can translate complex physiological signals into clinical narratives, it is unclear how agentic systems perform relative to zero-shot inference. To address these questions, we present Vivaldi, a role-structured multi-agent system that explains multivariate physiological time series. Due to regulatory constraints that preclude live deployment, we instantiate Vivaldi in a controlled, clinical pilot to a small, highly qualified cohort of emergency medicine experts, whose evaluations reveal a context-dependent picture that contrasts with prevailing assumptions that agentic reasoning uniformly improves performance. Our experiments show that agentic pipelines substantially benefit non-thinking and medically fine-tuned models, improving expert-rated explanation justification and relevance by +6.9 and +9.7 points, respectively. Contrarily, for thinking models, agentic orchestration often degrades explanation quality, including a 14-point drop in relevance, while improving diagnostic precision (ESI F1 +3.6). We also find that explicit tool-based computation is decisive for codifiable clinical metrics, whereas subjective targets, such as pain scores and length of stay, show limited or inconsistent changes. Expert evaluation further indicates that gains in clinical utility depend on visualization conventions, with medically specialized models achieving the most favorable trade-offs between utility and clarity. Together, these findings show that the value of agentic AI lies in the selective externalization of computation and structure rather than in maximal reasoning complexity, and highlight concrete design trade-offs and learned lessons, broadly applicable to explainable AI in safety-critical healthcare settings.

Davide Gabrielli, Paola Velardi, Stefano Faralli +1

Sapienza University of Rome Rome, Italy · ISTC-CNR & Sapienza University of Rome Rome, Italy · Technical University of Munich Munich, Germany

Explainable Artificial Intelligence Multi-Agent LLMs AI Agent Evaluation

Mar 4, 2026

A Rubric-Supervised Critic from Sparse Real-World Outcomes

Academic benchmarks for coding agents tend to reward autonomous task completion, measured by verifiable rewards such as unit-test success. In contrast, real-world coding agents operate with humans in the loop, where success signals are typically noisy, delayed, and sparse. How can we bridge this gap? In this paper, we propose a process to learn a "critic" model from sparse and noisy interaction data, which can then be used both as a reward model for either RL-based training or inference-time scaling. Specifically, we introduce Critic Rubrics, a rubric-based supervision framework with 24 behavioral features that can be derived from human-agent interaction traces alone. Using a semi-supervised objective, we can then jointly predict these rubrics and sparse human feedback (when present). In experiments, we demonstrate that, despite being trained primarily from trace-observable rubrics and sparse real-world outcome proxies, these critics improve best-of-N reranking on SWE-bench (Best@8 +15.9 over Random@8 over the rerankable subset of trajectories), enable early stopping (+17.7 with 83% fewer attempts), and support training-time data curation via critic-selected trajectories.

Xingyao Wang, Valerie Chen, Heng Ji +1

OpenHands · UIUC · CMU

Reward Modeling AI Agent Evaluation RL for Code Generation

Mar 3, 2026

Asymmetric Goal Drift in Coding Agents Under Value Conflict

Coding agents are increasingly deployed autonomously, at scale, and over long-context horizons. To be effective and safe, these agents must navigate complex trade-offs in deployment, balancing influence from the user, their learned values, and the codebase itself. Understanding how agents resolve these trade-offs in practice is critical, yet prior work has relied on static, synthetic settings that do not capture the complexity of real-world environments. To this end, we introduce a framework built on OpenCode in which a coding agent completes realistic, multi-step tasks under a system prompt constraint favoring one side of a value trade-off. We measure how often the agent violates this constraint as it completes tasks, with and without environmental pressure toward the competing value. Using this framework, we demonstrate that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit

\textit{asymmetric drift}

: they are more likely to violate their system prompt when its constraint opposes strongly-held values like security and privacy. We find for the models and values tested that goal drift correlates with three compounding factors: value alignment, adversarial pressure, and accumulated context. However, even constraints aligned with strongly-held values like privacy are violated under sustained environmental pressure for some models. Our findings reveal that shallow compliance checks are insufficient, and that environmental signals can override explicit constraints in ways that appear exploitable. Malicious actors with access to the codebase could manipulate agent behavior by appealing to learned values, with the risk compounding over the long horizons typical of agentic deployment.

Magnus Saebo, Spencer Gibson, Tyler Crosse +3

Columbia University · Independent · Georgia Tech +3

AI Coding Agents AI Agent Safety AI Agent Reliability AI Agent Evaluation

Mar 3, 2026

AgentAssay: Token-Efficient Regression Testing for Non-Deterministic AI Agent Workflows

Autonomous AI agents are deployed at unprecedented scale, yet no principled methodology exists for verifying that an agent has not regressed after changes to its prompts, tools, models, or orchestration logic. We present AgentAssay, the first token-efficient framework for regression testing non-deterministic AI agent workflows, achieving 78-100% cost reduction while maintaining rigorous statistical guarantees. Our contributions include: (1) stochastic three-valued verdicts (PASS/FAIL/INCONCLUSIVE) grounded in hypothesis testing; (2) five-dimensional agent coverage metrics; (3) agent-specific mutation testing operators; (4) metamorphic relations for agent workflows; (5) CI/CD deployment gates as statistical decision procedures; (6) behavioral fingerprinting that maps execution traces to compact vectors, enabling multivariate regression detection; (7) adaptive budget optimization calibrating trial counts to behavioral variance; and (8) trace-first offline analysis enabling zero-cost testing on production traces. Experiments across 5 models (GPT-5.2, Claude Sonnet 4.6, Mistral-Large-3, Llama-4-Maverick, Phi-4), 3 scenarios, and 7,605 trials demonstrate that behavioral fingerprinting achieves 86% detection power where binary testing has 0%, SPRT reduces trials by 78%, and the full pipeline achieves 100% cost savings through trace-first analysis. Implementation: 20,000+ lines of Python, 751 tests, 10 framework adapters.

Varun Pratap Bhardwaj

Independent Researcher

AI Agent Evaluation Mutation Testing

Mar 3, 2026

LiveAgentBench: Comprehensive Benchmarking of Agentic Systems Across 104 Real-World Challenges

As large language models grow more capable, general AI agents have become increasingly prevalent in practical applications. However, existing benchmarks face significant limitations, failing to represent real-world user tasks accurately. To address this gap, we present LiveAgentBench, a comprehensive benchmark with 104 scenarios that reflect real user requirements. It is constructed from publicly sourced questions on social media and real-world products. Central to our approach is the Social Perception-Driven Data Generation (SPDG) method, a novel process we developed to ensure each question's real-world relevance, task complexity, and result verifiability. We evaluate various models, frameworks, and commercial products using LiveAgentBench, revealing their practical performance and identifying areas for improvement. This release includes 374 tasks, with 125 for validation and 249 for testing. The SPDG process enables continuous updates with fresh queries from real-world interactions.

Hao Li, Huan Wang, Jinjie Gu +3

Ant Group

AI Agent Evaluation AI Agent Benchmarks Benchmark Construction

Mar 1, 2026

MOSAIC: A Universal Agent-Level Interface for Cross-Paradigm Agent Mixing and Human-AI Collaboration

Existing infrastructure cannot deploy agents from different decision-making paradigms within the same environment, making fair cross-paradigm comparison under identical conditions impossible. We present MOSAIC, an open-source platform that enables heterogeneous agents (RL policies, LLMs, VLMs, and human operators) to act within shared reinforcement learning environments in ad-hoc team settings with reproducible results. MOSAIC introduces three contributions. (i) IPC-based worker protocol that wraps native and third-party frameworks as isolated subprocess workers, each executing its own training and inference logic unmodified and communicating through a versioned inter-process protocol. (ii) An operator abstraction that forms an agent-level interface by mapping workers to agent slots: each operator, regardless of whether it is backed by an RL policy, an LLM, or a human, conforms to a minimal universal interface. (iii) A deterministic cross-paradigm evaluation framework with two complementary modes: a manual mode that advances up to

N

operators in lock-step under shared seeds for fine-grained visual inspection of behavioural differences; and a script mode that drives automated, long-running evaluation via declarative Python scripts for reproducible experiments. Our documentation is released at: https://mosaic-platform.readthedocs.io.

Abdulhamid M. Mousa, Jinhui Pang, Rakhmonberdi Khajiev +5

School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China · School of Computer Science and Technology, Beijing Institute of Technology, Beijing 100081, China · Mathematics Department, Faculty of Science, Ain Shams University, Cairo, Egypt +1

AI Agent Evaluation Agent Evaluation Multi-Agent Collaboration Human-in-the-Loop AI

Mar 1, 2026

How Well Does Agent Development Reflect Real-World Work?

AI agents are increasingly developed and evaluated on benchmarks relevant to human work, yet it remains unclear how representative these benchmarking efforts are of the labor market as a whole. In this work, we systematically study the relationship between agent development efforts and the distribution of real-world human work by mapping benchmark instances to work domains and skills. We first analyze 43 benchmarks and 72,342 tasks, measuring their alignment with human employment and capital allocation across all 1,016 real-world occupations in the U.S. labor market. We reveal substantial mismatches between agent development that tends to be programming-centric, and the categories in which human labor and economic value are concentrated. Within work areas that agents currently target, we further characterize current agent utility by measuring their autonomy levels, providing practical guidance for agent interaction strategies across work scenarios. Building on these findings, we propose three measurable principles for designing benchmarks that better capture socially important and technically challenging forms of work: coverage, realism, and granular evaluation.

Zora Zhiruo Wang, Sanidhya Vijayvargiya, Aspen Chen +7

AI Agent Benchmarks AI and Labor AI Agent Evaluation

Feb 28, 2026

COMBAT: Conditional World Models for Behavioral Agent Training

Recent advances in video generation have spurred the development of world models capable of simulating 3D-consistent environments and interactions with static objects. However, a significant limitation remains in their ability to model dynamic, reactive agents that can intelligently influence and interact with the world. To address this gap, we introduce COMBAT, a real-time, action-controlled world model trained on the complex 1v1 fighting game Tekken 3. Our work demonstrates that diffusion models can successfully simulate a dynamic opponent that reacts to player actions, learning its behavior implicitly. Our approach utilizes a 1.2 billion parameter Diffusion Transformer, conditioned on latent representations from a deep compression autoencoder. We employ state-of-the-art techniques, including causal distillation and diffusion forcing, to achieve real-time inference. Crucially, we observe the emergence of sophisticated agent behavior by training the model solely on single-player inputs, without any explicit supervision for the opponent's policy. Unlike traditional imitation learning methods, which require complete action labels, COMBAT learns effectively from partially observed data to generate responsive behaviors for a controllable Player 1. We present an extensive study and introduce novel evaluation methods to benchmark this emergent agent behavior, establishing a strong foundation for training interactive agents within diffusion-based world models.

Anmol Agarwal, Pranay Meshram, Sumer Singh +5

Overworld AI · Indian Institute of Science Education and Research Bhopal

Action-Conditioned World Models World Model Learning Game-Playing Agents AI Agent Evaluation

Feb 27, 2026

TraderBench: How Robust Are AI Agents in Adversarial Capital Markets?

Evaluating AI agents in finance faces two key challenges: static benchmarks require costly expert annotation yet miss the dynamic decision-making central to real-world trading, while LLM-based judges introduce uncontrolled variance on domain-specific tasks. We introduce TraderBench, a benchmark that addresses both issues. It combines expert-verified static tasks (knowledge retrieval, analytical reasoning) with adversarial trading simulations scored purely on realized performance-Sharpe ratio, returns, and drawdown-eliminating judge variance entirely. The framework features two novel tracks: crypto trading with four progressive market-manipulation transforms, and options derivatives scoring across P&L accuracy, Greeks, and risk management. Trading scenarios can be refreshed with new market data to prevent benchmark contamination. Evaluating 13 models (8B open-source to frontier) on ~50 tasks, we find: (1) 8 of 13 models score ~33 on crypto with <1-point variation across adversarial conditions, exposing fixed non-adaptive strategies; (2) extended thinking helps retrieval (+26 points) but has zero impact on trading (+0.3 crypto, -0.1 options). These findings reveal that current agents lack genuine market adaptation, underscoring the need for performance-grounded evaluation in finance.

Xiaochuang Yuan, Hui Xu, Silvia Xu +2

UC Santa Cruz

AI Agent Evaluation AI Agent Benchmarks Adversarial Robustness Algorithmic Trading Quantitative Finance

Feb 27, 2026

A Minimal Agent for Automated Theorem Proving

We propose a minimal agentic baseline that enables systematic comparison across different AI-based theorem prover architectures. This design implements the core features shared among state-of-the-art systems: iterative proof refinement, library search and context management. We evaluate this agentic approach using qualitatively different benchmarks and compare various frontier language models and design choices. Our results show competitive performance compared to state-of-the-art approaches, while using a significantly simpler architecture and a fraction of their cost. Additionally, we demonstrate consistent advantages of an iterative approach over multiple single-shot generations, especially in terms of sample efficiency and cost effectiveness. The implementation is released open-source as a candidate reference for future research and as an accessible prover for the community.

Borja Requena, Austin Letson, Krystian Nowakowski +2

Axiomatic AI, Barcelona, Spain.

Large Language Model-Guided Theorem Proving Automated Theorem Proving AI Agent Evaluation Iterative Refinement

Feb 26, 2026

Perfect score on IPhO 2025 theory by Gemini agent

The International Physics Olympiad (IPhO) is the world's most prestigious and renowned physics competition for pre-university students. IPhO problems require complex reasoning based on deep understanding of physical principles in a standard general physics curriculum. On IPhO 2025 theory problems, while gold medal performance by AI models was reported previously, it falls behind the best human contestant. Here we build a simple agent with Gemini 3.1 Pro Preview. We run it five times and it achieved a perfect score every time. However, data contamination could occur because Gemini 3.1 Pro Preview was released after the competition.

Yichen Huang

Santa Clarita, California 91350, USA

AI Agent Evaluation AI Agent Benchmarks Physical Reasoning LLM Reasoning

Feb 26, 2026

Evaluating Stochasticity in Deep Research Agents

Deep Research Agents (DRAs) are promising agentic systems that gather and synthesize information to support research across domains such as financial decision-making, medical analysis, and scientific discovery. Despite recent improvements in research quality (e.g., outcome accuracy when ground truth is available), DRA system design often overlooks a critical barrier to real-world deployment: stochasticity. Under identical queries, repeated executions of DRAs can exhibit substantial variability in terms of research outcome, findings, and citations. In this paper, we formalize the study of stochasticity in DRAs by modeling them as information acquisition Markov Decision Processes. We introduce an evaluation framework that quantifies variance in the system and identify three sources of it: information acquisition, information compression, and inference. Through controlled experiments, we investigate how stochasticity from these modules across different decision steps influences the variance of DRA outputs. Our results show that reducing stochasticity can improve research output quality, with inference and early-stage stochasticity contributing the most to DRA output variance. Based on these findings, we propose strategies for mitigating stochasticity while maintaining output quality via structured output and ensemble-based query generation. Our experiments on DeepSearchQA show that our proposed mitigation methods reduce average stochasticity by 22% while maintaining high research quality.

Haotian Zhai, Elias Stengel-Eskin, Pratik Patil +1

University of Texas at Austin∗

Deep Research Agents AI Agent Reliability AI Agent Evaluation

Feb 26, 2026

SkillNet: Create, Evaluate, and Connect AI Skills

Current AI agents can flexibly invoke tools and execute complex tasks, yet their long-term advancement is hindered by the lack of systematic accumulation and transfer of skills. Without a unified mechanism for skill consolidation, agents frequently ``reinvent the wheel'', rediscovering solutions in isolated contexts without leveraging prior strategies. To address this challenge, we introduce SkillNet, an open infrastructure for creating, evaluating, and organizing AI skills at scale. SkillNet structures skills within a unified ontology that supports creating skills from heterogeneous sources, establishing rich relational connections, and performing multi-dimensional evaluation across Safety, Completeness, Executability, Maintainability, and Cost-awareness. Our infrastructure integrates a repository of over 600,000 skills, an interactive platform, and a versatile Python toolkit. Experiments on ALFWorld, WebShop, and ScienceWorld show 40% higher average rewards and 30% fewer execution steps across multiple backbone models. Furthermore, SkillNet-Gym benchmarks skill retrieval, utilization, and composition, while SkillNet-Fabric enables task-specific skill routing through lightweight Wikis. By formalizing skills as evolving, composable assets, SkillNet provides a robust foundation for agents to move from transient experience to durable mastery.

Yuan Liang, Ruobin Zhong, Haoming Xu +47

LLM Agent Skill Evaluation Agent Skill Learning LLM Agent Skill Retrieval AI Agent Evaluation AI Agent Benchmarks

Feb 26, 2026

AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors

We introduce AuditBench, an alignment auditing benchmark. AuditBench consists of 56 language models with implanted hidden behaviors. Each model has one of 14 concerning behaviors--such as sycophantic deference, opposition to AI regulation, or secret geopolitical loyalties--which it does not confess to when directly asked. AuditBench models are highly diverse--some are subtle, while others are overt, and we use varying training techniques both for implanting behaviors and training models not to confess. To demonstrate AuditBench's utility, we develop an investigator agent that autonomously employs a configurable set of auditing tools. By measuring investigator agent success using different tools, we can evaluate their efficacy. Notably, we observe a tool-to-agent gap, where tools that perform well in standalone non-agentic evaluations fail to translate into improved performance when used with our investigator agent. We find that our most effective tools involve scaffolded calls to auxiliary models that generate diverse prompts for the target. White-box interpretability tools can be helpful, but the agent performs best with black-box tools. We also find that audit success varies greatly across training techniques: models trained on synthetic documents are easier to audit than models trained on demonstrations, with better adversarial training further increasing auditing difficulty. We release our models, agent, and evaluation framework to support future quantitative, iterative science on alignment auditing.

Abhay Sheshadri, Aidan Ewart, Elias Kempf +8

Anthropic Fellows Program · Anthropic

Benchmark Auditing AI Agent Evaluation Language Model Auditing

Feb 25, 2026

A Framework for Assessing AI Agent Decisions and Outcomes in AutoML Pipelines

Agent-based AutoML systems rely on large language models to make complex, multi-stage decisions across data processing, model selection, and evaluation. However, existing evaluation practices remain outcome-centric, focusing primarily on final task performance. Through a review of prior work, we find that none of the surveyed agentic AutoML systems report structured, decision-level evaluation metrics intended for post-hoc assessment of intermediate decision quality. To address this limitation, we propose an Evaluation Agent (EA) that performs decision-centric assessment of AutoML agents without interfering with their execution. The EA is designed as an observer that evaluates intermediate decisions along four dimensions: decision validity, reasoning consistency, model quality risks beyond accuracy, and counterfactual decision impact. Across four proof-of-concept experiments, we demonstrate that the EA can (i) detect faulty decisions with an F1 score of 0.919, (ii) identify reasoning inconsistencies independent of final outcomes, and (iii) attribute downstream performance changes to agent decisions, revealing impacts ranging from -4.9% to +8.3% in final metrics. These results illustrate how decision-centric evaluation exposes failure modes that are invisible to outcome-only metrics. Our work reframes the evaluation of agentic AutoML systems from an outcome-based perspective to one that audits agent decisions, offering a foundation for reliable, interpretable, and governable autonomous ML systems.

Gaoyuan Du, Amit Ahlawat, Xiaoyang Liu +1

Amazon Web Services · Amazon Security

AI Agent Evaluation AI Agent Auditing Automated Machine Learning LLM Agent Evaluation

Feb 24, 2026

Aletheia tackles FirstProof autonomously

We report the performance of Aletheia (Feng et al., 2026b), a mathematics research agent powered by Gemini 3 Deep Think, on the inaugural FirstProof challenge. Within the allowed timeframe of the challenge, Aletheia autonomously solved 6 problems (2, 5, 7, 8, 9, 10) out of 10 according to majority expert assessments; we note that experts were not unanimous on Problem 8 (only). For full transparency, we explain our interpretation of FirstProof and disclose details about our experiments as well as our evaluation. Raw prompts and outputs are available at https://github.com/google-deepmind/superhuman/tree/main/aletheia.

Tony Feng, Junehyuk Jung, Sang-hyun Kim +14

Google DeepMind · UC Berkeley · Brown University +7

LLM Mathematical Reasoning Autonomous Research Agents AI Agent Evaluation

Feb 23, 2026

Implicit Intelligence -- Evaluating Agents on What Users Don't Say

Real-world requests to AI agents are fundamentally underspecified. Natural human communication relies on shared context and unstated constraints that speakers expect listeners to infer. Current agentic benchmarks test explicit instruction-following but fail to evaluate whether agents can reason about implicit requirements spanning accessibility needs, privacy boundaries, catastrophic risks, and contextual constraints. We present Implicit Intelligence, an evaluation framework testing whether AI agents can move beyond prompt-following to become genuine goal-fulfillers, paired with Agent-as-a-World (AaW), a harness where interactive worlds are defined in human-readable YAML files and simulated by language models. Our scenarios feature apparent simplicity in user requests, hidden complexity in correct solutions, and discoverability of constraints through environmental exploration. Evaluating 16 frontier and open-weight models across 205 scenarios, we find that even the best-performing model achieves only 48.3% scenario pass rate, revealing substantial room for improvement in bridging the gap between literal instruction-following and human-like contextual reasoning.

Ved Sirdeshmukh, Marc Wetter

Applied Machine Learning Research

AI Agent Evaluation AI Agent Benchmarks Implicit Reasoning in Language Models LLM Agent Harnesses

Feb 23, 2026

Quantifying the Expectation-Realisation Gap for Agentic AI Systems

Agentic AI systems are deployed with expectations of substantial productivity gains, yet rigorous empirical evidence reveals systematic discrepancies between pre-deployment expectations and post-deployment outcomes. We review controlled trials and independent validations across software engineering, clinical documentation, and clinical decision support to quantify this expectation-realisation gap. In software development, experienced developers expected a 24% speedup from AI tools but were slowed by 19% -- a 43 percentage-point calibration error. In clinical documentation, vendor claims of multi-minute time savings contrast with measured reductions of less than one minute per note, and one widely deployed tool showed no statistically significant effect. In clinical decision support, externally validated performance falls substantially below developer-reported metrics. These shortfalls are driven by workflow integration friction, verification burden, measurement construct mismatches, and systematic variation in who benefits and who does not. The evidence motivates structured planning frameworks that require explicit, quantified benefit expectations with human oversight costs factored in.

Sebastian Lobentanzer

Institute of Computational Biology, Computational Health Center, Helmholtz Center, Munich, Germany · German Center for Diabetes Research, Munich, Germany

AI Agent Evaluation Human Oversight of AI Software Engineering Agents

Feb 23, 2026

MAS-FIRE: Fault Injection and Reliability Evaluation for LLM-Based Multi-Agent Systems

As LLM-based Multi-Agent Systems (MAS) are increasingly deployed for complex tasks, ensuring their reliability has become a pressing challenge. Since MAS coordinate through unstructured natural language rather than rigid protocols, they are prone to semantic failures (e.g., hallucinations, misinterpreted instructions, and reasoning drift) that propagate silently without raising runtime exceptions. Prevailing evaluation approaches, which measure only end-to-end task success, offer limited insight into how these failures arise or how effectively agents recover from them. To bridge this gap, we propose MAS-FIRE, a systematic framework for fault injection and reliability evaluation of MAS. We define a taxonomy of 15 fault types covering intra-agent cognitive errors and inter-agent coordination failures, and inject them via three non-invasive mechanisms: prompt modification, response rewriting, and message routing manipulation. Applying MAS-FIRE to three representative MAS architectures, we uncover a rich set of fault-tolerant behaviors that we organize into four tiers: mechanism, rule, prompt, and reasoning. This tiered view enables fine-grained diagnosis of where and why systems succeed or fail. Our findings reveal that stronger foundation models do not uniformly improve robustness. We further show that architectural topology plays an equally decisive role, with iterative, closed-loop designs neutralizing over 40% of faults that cause catastrophic collapse in linear workflows. MAS-FIRE provides the process-level observability and actionable guidance needed to systematically improve multi-agent systems.

Jin Jia, Zhiling Deng, Zhuangbin Chen +2

Sun Yat-sen University Zhuhai, Guangdong, China

Multi-Agent LLM Systems LLM Agent Reliability AI Agent Evaluation Agent Failure Analysis

Feb 22, 2026

Anatomy of Agentic Memory: Taxonomy and Empirical Analysis of Evaluation and System Limitations

Agentic memory systems enable large language model (LLM) agents to maintain state across long interactions, supporting long-horizon reasoning and personalization beyond fixed context windows. Despite rapid architectural development, the empirical foundations of these systems remain fragile: existing benchmarks are often underscaled, evaluation metrics are misaligned with semantic utility, performance varies significantly across backbone models, and system-level costs are frequently overlooked. This survey presents a structured analysis of agentic memory from both architectural and system perspectives. We first introduce a concise taxonomy of MAG systems based on four memory structures. Then, we analyze key pain points limiting current systems, including benchmark saturation effects, metric validity and judge sensitivity, backbone-dependent accuracy, and the latency and throughput overhead introduced by memory maintenance. By connecting the memory structure to empirical limitations, this survey clarifies why current agentic memory systems often underperform their theoretical promise and outlines directions for more reliable evaluation and scalable system design.

Dongming Jiang, Yi Li, Songtao Wei +8

University of Texas at Dallas · University of California, Davis · Texas A&M University

Agent Memory Agent Memory Management AI Agent Evaluation Memory Agent Benchmarks