The BrowserGym Ecosystem for Web Agent Research

TL;DR

This paper introduces an extended BrowserGym ecosystem for web agent research, unifying benchmarks and providing tools for standardized evaluation, demonstrated through large-scale experiments comparing top LLMs across multiple web benchmarks.

Contribution

It extends BrowserGym to unify existing benchmarks and includes AgentLab for agent development, testing, and analysis, enabling consistent evaluation and comprehensive experiment management.

Findings

Claude-3.5-Sonnet outperforms others on most benchmarks

GPT-4o excels in vision-related tasks

Building robust web agents remains challenging due to environment complexity

Abstract

The BrowserGym ecosystem addresses the growing need for efficient evaluation and benchmarking of web agents, particularly those leveraging automation and Large Language Models (LLMs). Many existing benchmarks suffer from fragmentation and inconsistent evaluation methodologies, making it challenging to achieve reliable comparisons and reproducible results. In an earlier work, Drouin et al. (2024) introduced BrowserGym which aims to solve this by providing a unified, gym-like environment with well-defined observation and action spaces, facilitating standardized evaluation across diverse benchmarks. We propose an extended BrowserGym-based ecosystem for web agent research, which unifies existing benchmarks from the literature and includes AgentLab, a complementary framework that aids in agent creation, testing, and analysis. Our proposed ecosystem offers flexibility for integrating new benchmarks while ensuring consistent evaluation and comprehensive experiment management. As a supporting evidence, we conduct the first large-scale, multi-benchmark web agent experiment and compare the performance of 6 state-of-the-art LLMs across 6 popular web agent benchmarks made available in BrowserGym. Among other findings, our results highlight a large discrepancy between OpenAI and Anthropic's latests models, with Claude-3.5-Sonnet leading the way on almost all benchmarks, except on vision-related tasks where GPT-4o is superior. Despite these advancements, our results emphasize that building robust and efficient web agents remains a significant challenge, due to the inherent complexity of real-world web environments and the limitations of current models.