OfficeQA Pro: An Enterprise Benchmark for End-to-End Grounded Reasoning
2026-03-09 • Artificial Intelligence
Artificial IntelligenceComputation and LanguageInformation Retrieval
AI summaryⓘ
The authors created OfficeQA Pro, a test to see how well AI can understand and reason using many different types of documents from nearly 100 years of U.S. Treasury Bulletins. The test has 133 questions that need careful reading and math using both text and tables. Popular AI models struggle a lot on this test, scoring very low even with extra help like web access. The authors found that giving AI a special organized version of the documents helps a bit, but overall the AI still makes many mistakes. Their work shows that current AI isn’t yet good enough to reliably understand complex, real-world documents at a professional level.
OfficeQA Promulti-document reasoningU.S. Treasury Bulletinsnumerical reasoningparametric knowledgestructured document representationai_parse_documentlarge language modelsretrieval strategytabular data
Authors
Krista Opsahl-Ong, Arnav Singhvi, Jasmine Collins, Ivan Zhou, Cindy Wang, Ashutosh Baheti, Owen Oertell, Jacob Portes, Sam Havens, Erich Elsen, Michael Bendersky, Matei Zaharia, Xing Chen
Abstract
We introduce OfficeQA Pro, a benchmark for evaluating AI agents on grounded, multi-document reasoning over a large and heterogeneous document corpus. The corpus consists of U.S. Treasury Bulletins spanning nearly 100 years, comprising 89,000 pages and over 26 million numerical values. OfficeQA Pro consists of 133 questions that require precise document parsing, retrieval, and analytical reasoning across both unstructured text and tabular data. Frontier LLMs including Claude Opus 4.6, GPT-5.4, and Gemini 3.1 Pro Preview achieve less than 5% accuracy on OfficeQA Pro when relying on parametric knowledge, and less than 12% with additional access to the web. When provided directly with the document corpus, frontier agents still struggle on over half of questions, scoring 34.1% on average. We find that providing agents with a structured document representation produced by Databricks' ai_parse_document yields a 16.1% average relative performance gain across agents. We conduct additional ablations to study the effects of model selection, table representation, retrieval strategy, and test-time scaling on performance. Despite these improvements, significant headroom remains before agents can be considered reliable at enterprise-grade grounded reasoning.