ExtractBench introduces a benchmark for measuring value accuracy, record completeness, grounding, and cost in schema-guided extraction tasks. The dataset includes 4,869 pages across 370 enterprise documents to test how well agents follow user-defined schemas.
HOW THIS AFFECTS YOU
●
builderThis provides specific metrics for measuring the reliability and cost-efficiency of your document processing agents.
●
researcherUse this to evaluate how well your extraction models handle complex, multi-domain business documents.