OfficeQA Pro exposes enterprise reasoning gap

This is the kind of benchmark that matters because it punishes shallow pattern matching and forces models to do actual document work. The low ceiling is the point: it shows how far enterprise-grade grounded reasoning still has to go.

• –OfficeQA Pro is harder than typical QA suites because questions span dense PDFs, tables, retrieval, and numerical extraction across decades of documents
• –The reported gains from structured parsing suggest preprocessing quality can matter almost as much as raw model choice
• –Frontier models still struggle even with web access, which is a strong signal that public-web retrieval is not enough for closed, document-heavy workflows
• –For builders, the lesson is to optimize the full pipeline: OCR, structure extraction, retrieval, and answer validation