StressWeb: Benchmark for Evaluating Web Agents in Simulated Environments

StressWeb is a benchmark dataset for evaluating web agents under realistic interaction settings. The dataset includes simulated web environments and task instructions for multi-step web tasks. It was released by HuggingSelf for research purposes, with a last recorded update on 2026-04-21.

Use Cases

• Benchmarking agent performance based on multi-step web tasks.
• Evaluating agent robustness based on interactions across different simulated web environments.
• Developing new agent architectures based on the provided task instructions and environments.

Strengths

• Designed specifically for evaluating agents under realistic interaction settings.
• Includes distinct components for environments and task instructions, suggesting a structured evaluation framework.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

HuggingSelf

Freshness

Last updated 2026-04-21 02:39:42; freshness should be verified.