EconWebArena

EconWebArena is a curated benchmark for evaluating large language model (LLM) agents on complex, multimodal economic tasks grounded in real-world web content. It features question-answering tasks that require navigating authoritative websites, interpreting structured and visual data, and extracting precise economic information.

Loading the Dataset

Load EconWebArena data with the following code:

from datasets import load_dataset, Features, Value

# Define features
features = Features({
    "id": Value("int32"),
    "website": Value("string"),
    "task": Value("string"),
    "task_url": Value("string"),
    "answer": Value("string"),
    "answer_key": Value("string"),
    "answer_url": Value("string"),
    "category": Value("string"),
    "seed": Value("bool"),
})

# Load dataset
data = load_dataset(
    "EconWebArena/EconWebArena",
    features=features,
)

# Add a derived goal column
def add_goal_column(example):
    example["goal"] = (
        f"{example['task']} "
        f'The answer must come from a URL containing "{example["answer_key"]}".'
    )
    return example

# Apply to splits
data = data.map(add_goal_column)

📚 Citation

If you find EconWebArena useful, please consider citing our work:

@article{liu2025econwebarena,
  title={EconWebArena: Benchmarking Autonomous Agents on Economic Tasks in Realistic Web Environments},
  author={Liu, Zefang and Quan, Yinzhu},
  journal={arXiv preprint arXiv:2506.08136},
  year={2025}
}

For more details, please visit our website.