OfficeQA Pro: Results as Published

OfficeQA Pro tests whether an AI agent can answer questions that need grounded, multi-document reasoning over an enterprise document corpus. The corpus is the U.S. Treasury Bulletins, about 89,000 pages spanning nearly 100 years and more than 26 million numerical values, so the agent has to parse documents, retrieve the right pages and calculate across prose and tables. The benchmark has 133 questions, written by a team led by Krista Opsahl-Ong. The printed metric is accuracy, as a percentage. The paper reports that frontier agents scored under 5% from parametric knowledge alone, under 12% with web access, and averaged 34.1% when given the document corpus.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.