WebArena-Infinity

Generating Browser Environments with Verifiable Tasks at Scale

WebArena-Infinity is a scalable approach for automatically generating realistic web environments paired with verifiable tasks, enabling robust training and evaluation of general-purpose browser agents. Instead of relying on labor-intensive manual construction, our approach uses a multi-agent system that coordinates coding agents with privileged access to environment internals and browser agents that interact through the UI—to iteratively generate, test, audit, and refine environments. Starting from real-world artifacts such as product manuals and workflows, the approach produces high-fidelity applications, difficulty-graded tasks, and reliable programmatic verifiers, all optimized for future reinforcement learning. The result is a cost-efficient, reproducible, and extensible benchmark suite that captures meaningful real-world complexity while remaining fully automated.

Quick Start: Explore Existing Environments

The repo ships with pre-built environments you can brow