OSWorld provides the first unified real-computer benchmark for open-ended multimodal agent tasks, exposing large performance gaps between humans and state-of-the-art LLM/VLM agents.
hub Canonical reference
VisualWebArena: Evaluating Multimodal Agents on Realistic Visual Web Tasks
Canonical reference. 90% of citing Pith papers cite this work as background.
47
Pith papers citing it
1
external citations · Pith
Background
90% of classified citations
abstract
Autonomous agents capable of planning, reasoning, and executing actions on the web offer a promising avenue for automating computer tasks. However, the majority of existing benchmarks primarily focus on text-based agents, neglecting many natural tasks that require visual information to effectively solve. Given that most computer interfaces cater to human perception, visual information often augments textual data in ways that text-only models struggle to harness effectively. To bridge this gap, we introduce VisualWebArena, a benchmark designed to assess the performance of multimodal web agents on realistic \textit{visually grounded tasks}. VisualWebArena comprises of a set of diverse and complex web-based tasks that evaluate various capabilities of autonomous multimodal agents. To perform on this benchmark, agents need to accurately process image-text inputs, interpret natural language instructions, and execute actions on websites to accomplish user-defined objectives. We conduct an extensive evaluation of state-of-the-art LLM-based autonomous agents, including several multimodal models. Through extensive quantitative and qualitative analysis, we identify several limitations of text-only LLM agents, and reveal gaps in the capabilities of state-of-the-art multimodal language agents. VisualWebArena provides a framework for evaluating multimodal autonomous language agents, and offers insights towards building stronger autonomous agents for the web. Our code, baseline models, and data is publicly available at https://jykoh.com/vwa.
hub tools
citation-role summary
background 9
dataset 1
citation-polarity summary
representative citing papers
Prismata cuts web-agent prompt-injection attack success from 85.5% to 0.7% via Biba-inspired DOM trust labeling and mechanical least-privilege confinement without site annotations.
SEATauBench is the first agent benchmark for SEA languages, finding that performance holds for language-only changes but degrades sharply with full domain localization.
The paper builds SOPBench showing frequent SOP violations in agentic browsers and introduces SOPGuard to enforce the policy with low overhead in BrowserOS.
WebChallenger introduces PageMem and three architecture mechanisms to achieve competitive web navigation with open-weight LLMs on WebArena, VisualWebArena, Online-Mind2Web, and WorkArena without fine-tuning or site adapters.
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.
ScaleWoB generates 100+ synthetic interactive GUI environments and 1000+ verifiable tasks as web pages, releasing a 120-task mobile benchmark where state-of-the-art agents achieve 27.92% success (17.82% on long-horizon tasks) versus 92.08% for humans, with synthetic results generalizing to real apps
VISTA is a new benchmark for end-to-end visual spec-to-web-app generation by LLM agents, featuring five prompt conditions, manual UI annotations, multi-metric evaluation, and results on four agent systems showing partial decoupling of visual and functional performance.
ShopGym introduces ShopArena to convert live storefronts into self-contained sandbox shops and ShopGuru to synthesize 224 benchmark tasks, with validation showing structural preservation and positive correlation of agent performance between synthetic and live shops.
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.
Vibe Code Bench evaluates AI models on building complete web applications from specs, with the best of 16 models achieving 61.8% accuracy on the test split using autonomous browser evaluation.
SecureWebArena is a new benchmark suite for holistic security evaluation of LVLM-based web agents using diverse simulated environments, attack taxonomies, and multi-layered failure analysis across reasoning, behavior, and outcomes.
AndroidWorld is a dynamic, reproducible Android benchmark that generates unlimited natural-language tasks for autonomous agents and shows current agents succeed on only 30.6 percent of them.
Controlled visual and instruction perturbations expose large, systematic spatial-reasoning and scale brittleness in 7B GUI grounding models that standard fixed-scene benchmarks miss.
WebRetriever is a benchmark of 800 websites and 1,550 tasks with an automated evaluator (NavEval) achieving ~91–97% human agreement, showing current web agents succeed on only 11–37% of realistic tasks across three evaluation protocols.
Rhetor automates rehearsed live web-app demos with segment-synchronized narration and real-time voice QA using cross-modal UI-plus-code features, a grounded scripter, rehearsal loops, and timing invariants, with case-study metrics on four applications.
PhoneBuddy combines real-app and mock-app RL after shared SFT, raising real-phone task success from 36.67% to 45.33% and AndroidWorld from 60.3% to 83.2%.
ChainWorld builds 347 chains from atomic OSWorld tasks and benchmarks four agents under single-turn and multi-turn protocols, reporting a maximum 31% completion rate with distinct failure profiles.
LifeSkill is a verifier-guided skill learning plus online internalization framework that raises average performance by 7 points over lifelong agent baselines on LifelongAgentBench.
SCALE introduces three adversarial roles (Selector, Predictor, Judger) and a graph exploration method (SCALE-Hop) to enable MLLM-based web agents to self-discover limitations and improve, backed by the SCALE-20k dataset from 19 websites.
Introduces DocOS benchmark to test GUI agents on proactively locating, comprehending, and executing instructions from online documentation in interactive web settings.
SaaS-Bench benchmark shows LLM-based agents achieve under 4% end-to-end success on 106 realistic professional tasks spanning 23 deployable SaaS platforms.
ReVision reduces token usage by 46% and improves success rate by 3% on OSWorld, WebTailBench, and AgentNetBench by removing redundant visual patches from 5-history trajectories with Qwen2.5-VL-7B.
MMTB is a new benchmark with 105 multimedia terminal tasks that shows how audio and video access changes agent performance and evidence use in executable workflows.
citing papers explorer
-
OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments
OSWorld provides the first unified real-computer benchmark for open-ended multimodal agent tasks, exposing large performance gaps between humans and state-of-the-art LLM/VLM agents.
-
Prismata: Confining Cross-Site Prompt Injection in Web Agents
Prismata cuts web-agent prompt-injection attack success from 85.5% to 0.7% via Biba-inspired DOM trust labeling and mechanical least-privilege confinement without site annotations.
-
SEATauBench: Adapting Tool-Agent-User Evaluation Into Low-Resource Southeast Asian Languages
SEATauBench is the first agent benchmark for SEA languages, finding that performance holds for language-only changes but degrades sharply with full domain localization.
-
Same-Origin Policy for Agentic Browsers
The paper builds SOPBench showing frequent SOP violations in agentic browsers and introduces SOPGuard to enforce the policy with low overhead in BrowserOS.
-
WebChallenger: A Reliable and Efficient Generalist Web Agent
WebChallenger introduces PageMem and three architecture mechanisms to achieve competitive web navigation with open-weight LLMs on WebArena, VisualWebArena, Online-Mind2Web, and WorkArena without fine-tuning or site adapters.
-
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.
-
ScaleWoB: Guiding GUI Agents with Coding Agents via Large-Scale Environmental Synthesis
ScaleWoB generates 100+ synthetic interactive GUI environments and 1000+ verifiable tasks as web pages, releasing a 120-task mobile benchmark where state-of-the-art agents achieve 27.92% success (17.82% on long-horizon tasks) versus 92.08% for humans, with synthetic results generalizing to real apps
-
VISTA: An End-to-End Benchmark for Visual Spec-to-Web-App Coding Agents
VISTA is a new benchmark for end-to-end visual spec-to-web-app generation by LLM agents, featuring five prompt conditions, manual UI annotations, multi-metric evaluation, and results on four agent systems showing partial decoupling of visual and functional performance.
-
ShopGym: An Integrated Framework for Realistic Simulation and Scalable Benchmarking of E-Commerce Web Agents
ShopGym introduces ShopArena to convert live storefronts into self-contained sandbox shops and ShopGuru to synthesize 224 benchmark tasks, with validation showing structural preservation and positive correlation of agent performance between synthetic and live shops.
-
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.
-
Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development
Vibe Code Bench evaluates AI models on building complete web applications from specs, with the best of 16 models achieving 61.8% accuracy on the test split using autonomous browser evaluation.
-
SecureWebArena: A Holistic Security Evaluation Benchmark for LVLM-based Web Agents
SecureWebArena is a new benchmark suite for holistic security evaluation of LVLM-based web agents using diverse simulated environments, attack taxonomies, and multi-layered failure analysis across reasoning, behavior, and outcomes.
-
AndroidWorld: A Dynamic Benchmarking Environment for Autonomous Agents
AndroidWorld is a dynamic, reproducible Android benchmark that generates unlimited natural-language tasks for autonomous agents and shows current agents succeed on only 30.6 percent of them.
-
GUI-Perturbed: Domain Randomization Reveals Systematic Brittleness in GUI Grounding Models
Controlled visual and instruction perturbations expose large, systematic spatial-reasoning and scale brittleness in 7B GUI grounding models that standard fixed-scene benchmarks miss.
-
WebRetriever: A Large-Scale Comprehensive Benchmark for Efficient Web Agent Evaluation
WebRetriever is a benchmark of 800 websites and 1,550 tasks with an automated evaluator (NavEval) achieving ~91–97% human agreement, showing current web agents succeed on only 11–37% of realistic tasks across three evaluation protocols.
-
Rehearsed Multi-Agent Live Product Demonstrations with Real-Time Voice Question Answering
Rhetor automates rehearsed live web-app demos with segment-synchronized narration and real-time voice QA using cross-modal UI-plus-code features, a grounded scripter, rehearsal loops, and timing invariants, with case-study metrics on four applications.
-
PhoneBuddy: Training Open Models for Agentic Phone Use
PhoneBuddy combines real-app and mock-app RL after shared SFT, raising real-phone task success from 36.67% to 45.33% and AndroidWorld from 60.3% to 83.2%.
-
ChainWorld: Composing Long-Horizon Desktop Workloads from Atomic OSWorld Tasks
ChainWorld builds 347 chains from atomic OSWorld tasks and benchmarks four agents under single-turn and multi-turn protocols, reporting a maximum 31% completion rate with distinct failure profiles.
-
Learning While Acting: A Skill-Enhanced Test-Time Co-Evolution Framework for Online Lifelong Learning Agents
LifeSkill is a verifier-guided skill learning plus online internalization framework that raises average performance by 7 points over lifelong agent baselines on LifelongAgentBench.
-
Learning to Adapt: Self-Improving Web Agent via Cognitive-Aware Exploration
SCALE introduces three adversarial roles (Selector, Predictor, Judger) and a graph exploration method (SCALE-Hop) to enable MLLM-based web agents to self-discover limitations and improve, backed by the SCALE-20k dataset from 19 websites.
-
DocOS: Towards Proactive Document-Guided Actions in GUI Agents
Introduces DocOS benchmark to test GUI agents on proactively locating, comprehending, and executing instructions from online documentation in interactive web settings.
-
SaaS-Bench: Can Computer-Use Agents Leverage Real-World SaaS to Solve Professional Workflows?
SaaS-Bench benchmark shows LLM-based agents achieve under 4% end-to-end success on 106 realistic professional tasks spanning 23 deployable SaaS platforms.
-
ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy Reduction
ReVision reduces token usage by 46% and improves success rate by 3% on OSWorld, WebTailBench, and AgentNetBench by removing redundant visual patches from 5-history trajectories with Qwen2.5-VL-7B.
-
MMTB: Evaluating Terminal Agents on Multimedia-File Tasks
MMTB is a new benchmark with 105 multimedia terminal tasks that shows how audio and video access changes agent performance and evidence use in executable workflows.
-
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.
-
AdaRubric: Task-Adaptive Rubrics for Reliable LLM Agent Evaluation and Reward Learning
AdaRubric adaptively generates task-specific rubrics via LLM, scores agent trajectories with per-dimension confidence weighting, and produces filtered DPO pairs that raise human correlation to Pearson r=0.79 and downstream task success by 6.8-8.5%.
-
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.
-
From Refusal to Recovery: A Control-Theoretic Approach to Generative AI Guardrails
Control-theoretic guardrails enable proactive correction of risky LLM agent actions in latent space, preventing catastrophes like collisions or bankruptcy while preserving task performance in simulated environments.
-
VS-Bench: Evaluating VLMs for Strategic Abilities in Multi-Agent Environments
VS-Bench is a new benchmark of ten visual multi-agent environments that measures VLMs on element recognition, next-action prediction, and normalized episode return, showing strong perception but large gaps in reasoning and decision-making with the best model at 46.6% prediction accuracy and 31.4% of
-
Grounded Reinforcement Learning for Visual Reasoning
ViGoRL introduces visually grounded RL that anchors reasoning steps to image coordinates and uses multi-turn zooming to outperform standard RL and supervised baselines on spatial and GUI reasoning benchmarks.
-
FieldWorkArena: Agentic AI Benchmark for Real Field Work Tasks
A benchmark with 886 real-world field work tasks across factories, warehouses, and retail, plus an evaluation method for multimodal AI agents.
-
LongVILA: Scaling Long-Context Visual Language Models for Long Videos
LongVILA scales visual-language models from 8 to 2048 video frames with 99.8% needle-in-a-haystack accuracy using long-context extension, supervised fine-tuning, and multi-modal sequence parallelism on up to 256 GPUs.
-
WebCanvas: Benchmarking Web Agents in Online Environments
WebCanvas creates a dynamic benchmark for web agents with a noise-resistant evaluation metric, the Mind2Web-Live dataset of 542 tasks, and open-source tools and agent framework for ongoing online testing.
-
OpenRath: Session-Centered Runtime State for Agent Systems
OpenRath introduces Session as a first-class, branchable runtime value that unifies fragmented state in multi-agent systems and makes fork, merge, and replay explicit operations.
-
CaptchaMind: Training CAPTCHA Solvers via Reinforcement Learning with Explicit Reasoning Supervision
Presents CaptchaBench benchmark and CaptchaMind RL solver achieving 82.9% success on benchmark tasks and 71% on real-world CAPTCHAs via explicit reasoning process supervision.
-
From Agent Loops to Deterministic Graphs: Execution Lineage for Reproducible AI-Native Work
Execution lineage models AI-native work as a DAG of computations with explicit dependencies, achieving perfect state preservation in controlled update tasks where loop-based agents introduce churn and contamination.
-
Tuning Qwen2.5-VL to Improve Its Web Interaction Skills
Two-stage fine-tuning of Qwen2.5-VL-32B improves success rates on single-click web tasks from 86% to 94%.
-
Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks
Plan-and-Act trains a dedicated Planner on synthetic plan-annotated trajectories to generate high-level plans that an Executor follows, reaching 57.58% success on WebArena-Lite and 81.36% on WebVoyager.
-
OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning
OPID distills episode- and step-level skills from completed on-policy trajectories, routes them via critical-first mechanism, and combines the resulting log-probability shift advantage with outcome advantage for policy optimization in language agents.
-
The Single-File Test: A Longitudinal Public-Interface Evaluation of First-Output LLM Web Generation with Social Reach Tracking
Claude outperformed other LLM families in generating functional single-file HTML under fixed public conditions, but neither technical variables nor prompt details reliably predicted 24-hour social media impressions.
-
Bridging Perception and Action: A Lightweight Multimodal Meta-Planner Framework for Robust Earth Observation Agents
The LMMP framework improves tool-calling accuracy and task success rates for Earth observation agents by grounding plans in multimodal features and remote sensing expert knowledge via a two-stage training process.
-
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 Reasoning for Large Language Models
The survey structures agentic reasoning for LLMs into foundational, self-evolving, and collective multi-agent layers while distinguishing in-context orchestration from post-training optimization and reviewing applications across domains.
-
Beyond NL2Code: A Structured Survey of Multimodal Code Intelligence
A structured survey of multimodal code intelligence that formulates the field by code roles and organizes work into four domains while proposing verification-centered research directions.
-
Toward Native Multimodal Modeling: A Roadmap
A roadmap that defines architectural nativity for multimodal models and categorizes them into Multi-to-Text, Multi-to-Target, and Multi-to-Multi types while outlining an industrial pipeline toward unified transformer-based native multimodal modeling.
-
LLM-Powered AI Agent Systems and Their Applications in Industry
A survey categorizing LLM-powered agent systems into software-based, physical, and hybrid types, covering industrial applications and challenges such as latency and security.
- Signal-Driven Observation for Long-Horizon Web Agents