FinanceComplexQA: Benchmarking Agentic Reasoning on Industrial-grade Financial Documents

Researchers introduced FinanceComplexQA, a benchmark for testing agentic reasoning on complex financial documents.

It includes a synthetic document generation skill, 2,000 financial documents, 6,000 QA pairs, and 2,026 open-ended tasks across 1,009 documents.

The benchmark supports bilingual evaluation and uses multiple metrics to assess reasoning, summarization, and numerical accuracy.

FinanceComplexQA offers a realistic, difficult testbed for improving AI systems used in financial analysis and RAG settings.