Herculean: An Agentic Benchmark for Financial Intelligence
Published in arXiv preprint · under review at NeurIPS, 2026
Status: preprint on arXiv, submitted to NeurIPS 2026 — under review, results pending. · My role: ran several of the key experiments.
Summary
Herculean is a benchmark for AI-agent capability in realistic financial workflows. It spans four domains — Trading, Hedging, Market Insights, and Auditing — each implemented as a standardized environment with domain-specific tools and explicit success criteria, so agents are judged on completing a workflow rather than answering a question.
Evaluating frontier agents shows a clear split: they are relatively strong on Trading and Market Insights, but degrade sharply on Hedging and Auditing, where success demands long-horizon coordination, consistent state tracking, and structured verification.
A large multi-institution collaboration; I am one of ~55 authors.
Recommended citation: Xueqing Peng, Zhuohan Xie, Yupeng Cao, et al. (including Yonghan Yang). (2026). "Herculean: An Agentic Benchmark for Financial Intelligence." arXiv:2605.14355. Under review at NeurIPS 2026.
Download Paper