WebArena-Infinity: Generating Browser Environments with Verifiable Tasks at Scale

🏛 Institutions: Duke University
📅 Date: March 2026
📑 Publisher: Blog Post
💻 Env: Web
🔑 Keywords: benchmark dataset environment synthesis verifiable rewards reinforcement learning WebArena WebArena-Infinity

TLDR

WebArena-Infinity automates the generation of high-authenticity web environments with verifiable tasks from static artifacts like user manuals, using a multi-agent pipeline of coding and browser-use agents. It produces 10 environments with 1,260 tasks and 2,070 trajectories. Agents achieve notably lower success rates than on manually built benchmarks, suggesting the generated tasks capture meaningful complexity.

Open paper Code Report issue