Skip to content

WebVoyager

An LMM-driven web agent tested on 15 real sites

Visit official site

What WebVoyager is

The agent is described as a large multimodal model driven web agent that completes user instructions by interacting with real websites.

What you can do with it

  • Complete user instructions end to end
  • Interact with real-world websites
  • Execute website tasks automatically

Evaluation notes

  • Tasks come from 15 widely used websites
  • A 55.7% task success rate is reported
  • Automatic evaluation shows 85.3% agreement

Pros & cons

✓ What we like

  • The agent interacts with real-world sites
  • Automatic evaluation shows 85.3% agreement with human judgment

! What to watch out for

  • The described setting is a research evaluation
  • Evaluation depends on GPT-4V understanding

FAQ

What success rate is reported?

The material states a 55.7% task success rate.

How is it evaluated?

An automatic protocol using GPT-4V is proposed.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review