FinanceHarness: Autonomous Financial Deep Research Framework

2026-07-30

AI Summary

Existing general-purpose deep research frameworks struggle to meet the demanding requirements of finance, particularly in domain-specific expertise, historical pattern analysis, and event forecasting, while also lacking verifiable benchmarks that prevent information leakage. To address these limitations, this work proposes an end-to-end automated deep research framework tailored for finance, integrating a hierarchical control architecture (FinanceHarness) and a temporally isolated evaluation benchmark (FinanceGym). The framework incorporates practitioner-guided workflows, finance-specific tools, and a reward modeling mechanism. Built upon open-source large language models, it achieves a significant performance gain—raising overall system scores from 25.3% to 32.4% under the same base model—and attains an expert validation pass rate of 82%, substantially outperforming mainstream models, which typically score below 40%.

Abstract

Powered by advances in LLMs and autonomous agents, deep research has become one of the most widely adopted agentic products. However, most deep research systems write general-purpose reports, which are inadequate for financial deep research. Financial research demands specialized knowledge to analyze historical patterns and forecast upcoming events. Automating financial deep research therefore requires both a layered harness to drive the research agent and a verifiable, point-in-time benchmark that prevents leakage of future information. We present FinanceHarness, a harness that runs finance-oriented tools and practitioner-guided workflows, automating financial deep research end to end: environment and data construction, the agent execution loop, and reward modeling. We further propose FinanceGym, comprising thesis-driven research questions and rubrics that combine pre-cutoff and post-cutoff criteria. Professional expert validation yields an 82% pass rate. Even leading LLMs and agents score below 40% on the rubrics, showing that FinanceGym is challenging and leaves substantial headroom. With the same open-weight backbone, FinanceHarness improves the overall rubric score from 25.3% to 32.4%. FinanceHarness is available at https://github.com/Yijia-Xiao/FinanceHarness.

Problem

• financial deep research
• autonomous agents
• future information leakage
• domain-specific benchmark
• LLM evaluation

Innovation

• FinanceHarness
• autonomous financial research
• FinanceGym
• point-in-time benchmark
• LLM agent workflow