On 108 long-horizon real-world computer workflows, frontier agents complete at most 20.6% of tasks and fail mainly by losing hidden state, not by basic GUI control.
hub Canonical reference
The BrowserGym Ecosystem for Web Agent Research
Canonical reference. 75% of citing Pith papers cite this work as background.
32
Pith papers citing it
Background
75% of classified citations
abstract
The BrowserGym ecosystem addresses the growing need for efficient evaluation and benchmarking of web agents, particularly those leveraging automation and Large Language Models (LLMs). Many existing benchmarks suffer from fragmentation and inconsistent evaluation methodologies, making it challenging to achieve reliable comparisons and reproducible results. In an earlier work, Drouin et al. (2024) introduced BrowserGym which aims to solve this by providing a unified, gym-like environment with well-defined observation and action spaces, facilitating standardized evaluation across diverse benchmarks. We propose an extended BrowserGym-based ecosystem for web agent research, which unifies existing benchmarks from the literature and includes AgentLab, a complementary framework that aids in agent creation, testing, and analysis. Our proposed ecosystem offers flexibility for integrating new benchmarks while ensuring consistent evaluation and comprehensive experiment management. As a supporting evidence, we conduct the first large-scale, multi-benchmark web agent experiment and compare the performance of 6 state-of-the-art LLMs across 6 popular web agent benchmarks made available in BrowserGym. Among other findings, our results highlight a large discrepancy between OpenAI and Anthropic's latests models, with Claude-3.5-Sonnet leading the way on almost all benchmarks, except on vision-related tasks where GPT-4o is superior. Despite these advancements, our results emphasize that building robust and efficient web agents remains a significant challenge, due to the inherent complexity of real-world web environments and the limitations of current models.
hub tools
citation-role summary
background 6
dataset 1
method 1
citation-polarity summary
representative citing papers
Attackers can force LLM guardrails into extended reasoning loops via optimized payloads, causing 13-63x token amplification and up to 148x latency in agent systems.
AgentBeats implements agentified evaluation of diverse AI agents through standardized interfaces, validated at scale in a five-month competition with 298 judges and 467 subjects plus a coding case study.
MAC-Bench is a new adversarial benchmark that converts legal texts into executable scenarios via the SERV pipeline to measure procedural compliance in multi-agent LLM systems using CSR and MG metrics.
SentinelBench is a new benchmark for time-evolving monitoring tasks in web environments, measuring task completion, reaction time, and resource use with baselines from three models and two harnesses.
Frontier browser agents show strong resistance to hand-crafted multi-step prompt injections (0/140 success), unlike coding agents (up to 100%), indicating domain-conditioned safety and that prior high ASR reports may not generalize.
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
Open 4B and 8B visual web agents achieve state-of-the-art results on browser benchmarks by predicting actions from screenshots and instructions, outperforming similar open models and some closed larger-model agents, with full release of data and code planned.
This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.
Analysis of 13 coding agent scaffolds at pinned commits yields a 12-dimension taxonomy showing five composable loop primitives, with 11 agents combining multiple primitives instead of using one fixed structure.
RAG, MCP, and NLWeb interfaces let LLM web agents achieve higher F1 scores (0.75-0.77 vs 0.67) and much lower token usage and runtime than HTML in controlled e-commerce tasks.
WebMall is the first offline multi-shop benchmark for evaluating LLM web agents on complex comparison shopping tasks across heterogeneous product data from multiple simulated e-shops.
EconWebArena is a new benchmark with 360 curated economic tasks across 82 authoritative websites for evaluating multimodal web agents on navigation, grounding, and data extraction.
A capability-driven benchmark of 400 bilingual real-world tasks shows current proactive agents fail >50% of the time, with framework architecture impacting performance more than base model choice.
Signal-Driven Observation decouples observation from action frequency in long-horizon web agents by invoking selective task-relevant DOM reads only on signals such as URL changes or action failures.
Structured schema.org metadata still gives dataset-retrieval agents a large precision advantage for machine-actionable data.
SimGym is a browser-based VLM agent framework that simulates A/B test outcomes on e-commerce storefronts with 77% directional agreement on add-to-cart shifts from real buyer traffic.
Agent benchmarks can report evidence-supported score bounds instead of single misleading success rates by adding a layer that checks required artifacts for outcome verification.
Claw-Eval-Live benchmark with 105 tasks shows no frontier LLM agent exceeds 66.7% success rate on evolving real-world workflows, with HR and multi-system tasks as persistent bottlenecks.
VLAA-GUI adds mandatory visual verifiers, multi-tier loop breakers, and on-demand search to GUI agents, reaching 77.5% on OSWorld and 61.0% on WindowsAgentArena with some models exceeding human performance.
WebSP-Eval shows multimodal web agents fail many real security and privacy browser tasks, with stateful UI elements like toggles causing over 45% of failures.
WebFactory is a fully automated RL pipeline that compresses LLM-encoded internet knowledge into grounded web agents, achieving performance comparable to human-annotated training but using synthetic data from only 10 websites.
MobiBench reaches near-human offline evaluation fidelity for mobile GUI agents by accepting any valid action at each step, and enables modular attribution of performance to agent components.
The paper proposes Amazon-Bench, a functionality-grounded benchmark for web agents in e-commerce that generates diverse task queries from webpage elements and evaluates both task performance and safety risks.
citing papers explorer
-
OSWorld 2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks
On 108 long-horizon real-world computer workflows, frontier agents complete at most 20.6% of tasks and fail mainly by losing hidden state, not by basic GUI control.
-
From Shield to Target: Denial-of-Service Attacks on LLM-Based Agent Guardrails
Attackers can force LLM guardrails into extended reasoning loops via optimized payloads, causing 13-63x token amplification and up to 148x latency in agent systems.
-
AgentBeats: Agentifying Agent Assessment for Openness, Standardization, and Reproducibility
AgentBeats implements agentified evaluation of diverse AI agents through standardized interfaces, validated at scale in a five-month competition with 298 judges and 467 subjects plus a coding case study.
-
Beyond Goodhart's Law: A Dynamic Benchmark for Evaluating Compliance in Multi-Agent Systems
MAC-Bench is a new adversarial benchmark that converts legal texts into executable scenarios via the SERV pipeline to measure procedural compliance in multi-agent LLM systems using CSR and MG metrics.
-
SentinelBench: A Benchmark for Long-Running Monitoring Agents
SentinelBench is a new benchmark for time-evolving monitoring tasks in web environments, measuring task completion, reaction time, and resource use with baselines from three models and two harnesses.
-
Domain-Conditioned Safety in Frontier Computer-Using Agents: A 793-Episode Browser Benchmark, a Coding-Domain Cross-Reference, and a Reproducibility Audit of Recent Red-Teaming
Frontier browser agents show strong resistance to hand-crafted multi-step prompt injections (0/140 success), unlike coding agents (up to 100%), indicating domain-conditioned safety and that prior high ASR reports may not generalize.
-
WebGameBench: Requirement-to-Application Evaluation for Coding Agents via Browser-Native Games
WebGameBench is a new benchmark that evaluates coding agents on building browser-native games from frozen specifications, with runtime browser evaluation showing best agents reach 76.9% usable rate but only 20.2% excellent rate.
-
MolmoWeb: Open Visual Web Agent and Open Data for the Open Web
Open 4B and 8B visual web agents achieve state-of-the-art results on browser benchmarks by predicting actions from screenshots and instructions, outperforming similar open models and some closed larger-model agents, with full release of data and code planned.
-
Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning
This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.
-
Inside the Scaffold: A Source-Code Taxonomy of Coding Agent Architectures
Analysis of 13 coding agent scaffolds at pinned commits yields a 12-dimension taxonomy showing five composable loop primitives, with 11 agents combining multiple primitives instead of using one fixed structure.
-
MCP vs RAG vs NLWeb vs HTML: A Comparison of the Effectiveness and Efficiency of Different Agent Interfaces to the Web (Technical Report)
RAG, MCP, and NLWeb interfaces let LLM web agents achieve higher F1 scores (0.75-0.77 vs 0.67) and much lower token usage and runtime than HTML in controlled e-commerce tasks.
-
WebMall -- A Multi-Shop Benchmark for Evaluating Web Agents
WebMall is the first offline multi-shop benchmark for evaluating LLM web agents on complex comparison shopping tasks across heterogeneous product data from multiple simulated e-shops.
-
EconWebArena: Benchmarking Autonomous Agents on Economic Tasks in Realistic Web Environments
EconWebArena is a new benchmark with 360 curated economic tasks across 82 authoritative websites for evaluating multimodal web agents on navigation, grounding, and data extraction.
-
UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks
A capability-driven benchmark of 400 bilingual real-world tasks shows current proactive agents fail >50% of the time, with framework architecture impacting performance more than base model choice.
-
Signal-Driven Observation for Long-Horizon Web Agents
Signal-Driven Observation decouples observation from action frequency in long-horizon web agents by invoking selective task-relevant DOM reads only on signals such as URL changes or action failures.
-
Do Data Agents Need Semantic Metadata? A Comparative Study in Agentic Data Retrieval
Structured schema.org metadata still gives dataset-retrieval agents a large precision advantage for machine-actionable data.
-
SimGym: A Framework for A/B Test Simulation in E-Commerce with Traffic-Grounded VLM Agents
SimGym is a browser-based VLM agent framework that simulates A/B test outcomes on e-commerce storefronts with 77% directional agreement on add-to-cart shifts from real buyer traffic.
-
Can Agent Benchmarks Support Their Scores? Evidence-Supported Bounds for Interactive-Agent Evaluation
Agent benchmarks can report evidence-supported score bounds instead of single misleading success rates by adding a layer that checks required artifacts for outcome verification.
-
Claw-Eval-Live: A Live Agent Benchmark for Evolving Real-World Workflows
Claw-Eval-Live benchmark with 105 tasks shows no frontier LLM agent exceeds 66.7% success rate on evolving real-world workflows, with HR and multi-system tasks as persistent bottlenecks.
-
VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation
VLAA-GUI adds mandatory visual verifiers, multi-tier loop breakers, and on-demand search to GUI agents, reaching 77.5% on OSWorld and 61.0% on WindowsAgentArena with some models exceeding human performance.
-
WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks
WebSP-Eval shows multimodal web agents fail many real security and privacy browser tasks, with stateful UI elements like toggles causing over 45% of failures.
-
WebFactory: Automated Compression of Foundational Language Intelligence into Grounded Web Agents
WebFactory is a fully automated RL pipeline that compresses LLM-encoded internet knowledge into grounded web agents, achieving performance comparable to human-annotated training but using synthetic data from only 10 websites.
-
MobiBench: Multi-Branch, Modular Benchmark for Mobile GUI Agents
MobiBench reaches near-human offline evaluation fidelity for mobile GUI agents by accepting any valid action at each step, and enables modular attribution of performance to agent components.
-
A Functionality-Grounded Benchmark for Evaluating Web Agents in E-commerce Domains
The paper proposes Amazon-Bench, a functionality-grounded benchmark for web agents in e-commerce that generates diverse task queries from webpage elements and evaluates both task performance and safety risks.
-
AliyunConsoleAgent: Training Web Agents in Real-World Cloud Environments via Distillation and Reinforcement Learning
AliyunConsoleAgent-32B reaches 63.52% success on a 278-task cloud console benchmark, closing to 1.82pp of frontier models at 92% lower cost via SFT distillation and GRPO RL.
-
ProjGuard: Safety Monitoring for Computer-Use Agents via Low-Dimensional Projections
ProjGuard monitors agent trajectories with low-dimensional projections to cut unsafe actions from 16% to 3% and raise task completion from 59% to 65% on OS-Harm.
-
GUI Agents with Reinforcement Learning: Toward Digital Inhabitants
The paper delivers the first comprehensive overview of RL for GUI agents, organizing methods into offline, online, and hybrid strategies while analyzing trends in rewards, efficiency, and deliberation to outline a future roadmap.
-
Combating the Memory Walls: Optimization Pathways for Long-Context Agentic LLM Inference
PLENA introduces a co-designed system with three optimization pathways for long-context agentic LLM inference, claiming up to 2.23x throughput over A100 and 4.04x energy efficiency.
-
From Question Answering to Task Completion: A Survey on Agent System and Harness Design
Survey framing LLM agents as model-plus-harness systems, decomposing harness responsibilities, mapping them to tasks, and highlighting open challenges in evaluation, safety, and co-evolution.
-
BaRA: Budget-constrained and Reliable Web Data Collection Agent
BaRA improves valid link discovery and multimodal artifact extraction in budget-constrained web data collection via BFS liveness checks, rule-based validation, and self-reflection.
-
ClawEnvKit: Automatic Environment Generation for Claw-Like Agents
EVT improves the RMT backbone by using Euclidean-distance attention decay and 1D token grouping, achieving 86.6% top-1 on ImageNet-1K at 384×384 resolution.
-
Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges
A survey that taxonomizes threats to agentic AI, reviews benchmarks and evaluation methods, discusses technical and governance defenses, and identifies open challenges.