OpenWebVoyager: Building Multimodal Web Agents via Iterative Real-World Exploration, Feedback and Optimization

OpenWebVoyager is a multimodal web agent that improves itself through repeated cycles of real-world exploration, feedback collection, and policy optimization. It starts from imitation learning, mines open-web trajectories, and shows stronger performance after each optimization round.

• SCAFFOLD: Self-Improving Web Agents via Recursive Parametric Skill Abstraction
  August 31, 2026 · arXiv
• Thinking vs. Doing: Agents that Reason by Scaling Test-Time Interaction
  June 9, 2025 · SEA @ NeurIPS 2025 (Oral)
• Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems
  July 17, 2024 · arXiv
• Large Language Models Can Self-Improve At Web Agent Tasks
  May 30, 2024 · arXiv
• Autonomous Evaluation and Refinement of Digital Agents
  April 9, 2024 · COLM 2024
• BlueLM-GUI Technical Report: A Real-Device-Centric Flywheel for Self-Improving Mobile GUI Agents
  September 11, 2026 · arXiv
• OSExpert: Computer-Use Agents Learning Professional Skills via Exploration
  March 9, 2026 · arXiv
• Compress to Focus: Efficient Coordinate Compression for Policy Optimization in Multi-Turn GUI Agents
  January 14, 2026 · arXiv
• UI-Genie: A Self-Improving Approach for Iteratively Boosting MLLM-based Mobile GUI Agents
  May 27, 2025 · NeurIPS 2025 (Poster)
• VRL-Bench: Benchmarking agents on computer control tasks under finite trial budgets
  September 11, 2026 · arXiv
• Monitoring Web Agents Without Internal Signals: Observable Trajectories and Key-Step Supervision
  September 2, 2026 · arXiv
• Discriminative World Models for Web Agents
  September 2, 2026 · arXiv
• Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection Guardrails
  September 2, 2026 · arXiv
• When and What to Teach: Budget-Aware Online Adaptation for Web Agents
  August 31, 2026 · arXiv
• Learning Simple Test-Time Environments for LLM Web Agents
  August 29, 2026 · arXiv
• WebMCP-Phalanx: Enforcing and Characterizing Trust Boundaries for Browser-Integrated LLM Agents
  August 25, 2026 · arXiv
• BrowserForge: Scaling Web Episode via Parallel Browser Sandboxes
  August 25, 2026 · arXiv
• Training Needs Trustworthy Worlds: Verified Synthetic Web Environments for Agent Learning
  August 22, 2026 · arXiv
• Lexical Coupling in GUI Element Grounding: Sentence Embeddings Track Labels across Mobile and Web
  August 22, 2026 · arXiv
• ComponentBench: Diagnosing Component-Level Failures in Computer-Use Agents
  August 18, 2026 · COLM 2026
• StepJack: Benchmarking Computer-Use Agent Safety Against Multi-Step Indirect Prompt Injection
  August 6, 2026 · arXiv
• Routing Is Least Learnable Where It Is Most Valuable: Bounds on Representation Routing for Web Agents
  August 6, 2026 · arXiv
• LoginTrap: Uncovering Task-Agnostic Phishing-Style Indirect Prompt Injection Attacks against LLM-based Web Agents
  August 5, 2026 · arXiv
• Qwen-CUA: Native Computer Use for (almost) Everything
  August 3, 2026 · arXiv