TL;DR

WebArena provides self-hosted, realistic websites and benchmark tasks for agents that must navigate interfaces, maintain state, and complete multi-step goals.

Why it matters

Agent progress depends on reproducible environments with verifiable task outcomes. WebArena supplies both while remaining closer to real web work than toy tasks.

Key findings

  1. 01

    Long-horizon web tasks expose planning and grounding failures.

  2. 02

    Self-hosted environments enable reproducible evaluation and future training loops.

Scaling dimensions

Models, methods & benchmarks

Benchmarks
WebArena

Topics

Read the original sourcearXiv