aiqbench analysis
This article introduces FinBalance, a new benchmark designed to test large language models on the complex task of multi-document accounting reconciliation rather than simple artifact extraction. The study reveals that contemporary LLMs struggle significantly with binding numerical entries to their supporting source documents and maintaining consistent aggregation across ledgers. By exposing these specific failure modes in financial reasoning tasks, this research highlights critical gaps between plausible-sounding outputs and mathematically accurate bookkeeping required for real-world applications.
FinBalance: A Multi-Document Accounting Reconciliation Benchmark
arXiv:2606.15949v1 Announce Type: new Abstract: Existing financial-NLP benchmarks mostly evaluate prepared artifacts such as filings, tables, or extracted values. Real accounting begins earlier: source documents must be reconciled into cited journal entries, aggregated into a balance sheet, and checked for contradictions. We introduce FinBalance, a multi-document accounting reconciliation benchmark built from source-document bundles across eight industries, three period types, and five difficulty levels. Human-authored business scenarios, accounting policies, tax/FX treatments, document schemas, distractors, and inconsistency templates are composed by a deterministic generator whose ledger produces journal entries,balance sheets, and 23 inconsistency-code labels. On a 710-record evaluation split, six contemporary LLMs reach at most 46% exact final-balance-sheet accuracy. Four models show a 26-41 pp gap between BS_exact, the model's reported balance sheet, and BS_recon, the balance sheet obtained by replaying its entries through our ledger. Models often recover numerically plausible entries but fail to bind them to supporting documents and aggregate them consistently. Citation-pressure prompting barely changes document-linking errors, while ledger-feedback ablations substantially improve reported balance sheets and expose inconsistency-detection trade-offs. Expert finance reviewers validate the benchmark design and labels.
Abstract:Existing financial-NLP benchmarks mostly evaluate prepared artifacts such as filings, tables, or extracted values. Real accounting begins earlier: source documents must be reconciled into cited journal entries, aggregated into a balance sheet, and checked for contradictions. We introduce FinBalance, a multi-document accounting reconciliation benchmark built from source-document bundles across eight industries, three period types, and five difficulty levels. Human-authored business scenarios, accounting policies, tax/FX treatments, document schemas, distractors, and inconsistency templates are composed by a deterministic generator whose ledger produces journal entries,balance sheets, and 23 inconsistency-code labels. On a 710-record evaluation split, six contemporary LLMs reach at most 46% exact final-balance-sheet accuracy. Four models show a 26-41 pp gap between BS_exact, the model's reported balance sheet, and BS_recon, the balance sheet obtained by replaying its entries through our ledger. Models often recover numerically plausible entries but fail to bind them to supporting documents and aggregate them consistently. Citation-pressure prompting barely changes document-linking errors, while ledger-feedback ablations substantially improve reported balance sheets and expose inconsistency-detection trade-offs. Expert finance reviewers validate the benchmark design and labels.
18 pages, 12 figures. Code and data: this https URL
Computation and Language (cs.CL)
arXiv:2606.15949 [cs.CL]
(or arXiv:2606.15949v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.15949 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 279 words
Related coverage

Adaptive Instructed-Retriever: Frontier-Quality Search at 2x Lower Latency
Databricks · Sep 9, 2026

AI Applications in Finance: A Practical Use Case Guide
Databricks · Jul 24, 2026

A Candid Abacus AI Review: The All-in-One AI Platform for Professionals & Enterprises
KDnuggets · Sep 10, 2026

Vertical Advantage: Transforming Industries with Lakebase and Agentic AI
Databricks · Aug 27, 2026