FinLongDocQA: Financial Numerical Reasoning Over Long Documents

A benchmark dataset for evaluating document-level numerical reasoning across single and multiple tables in financial annual reports. The dataset was created by Amian and was last updated on the platform in April 2026. It requires models to retrieve relevant tables and text from full annual reports before computing answers to questions.

Use Cases

• Benchmarking model performance on financial numerical reasoning based on the described QA task.
• Training retrieval-augmented models for long financial documents based on the described need to find relevant tables and text.
• Developing multi-step reasoning systems for financial reports based on the cross-table and document-level analysis described.

Strengths

• Focuses on a specific, complex task of numerical reasoning over long financial documents.
• Provides a benchmark for evaluating both retrieval and reasoning capabilities, as described.

Limitations

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

Provenance

Source

Author: Amian. Source data appears to be financial annual reports.

Collection Method

Likely constructed as a benchmark with curated question-answer pairs based on financial reports.

Freshness

Last updated 2026-04-15 15:10:29; freshness should be verified.