Introducing ExtractBench: The Most Comprehensive Benchmark for Data Extraction from Enterprise Docs

LlamaIndex is open sourcing the first schema-guided document extraction benchmark for the agentic era, ExtractBench. Document extraction is the foundation of every AI agent that operates on real-world enterprise files. But as agents gain autonomy and humans step out of the loop, existing benchmarks fall short of measuring extraction quality across the requirements production actually demands. ExtractBench is an open-source benchmark testing 370 enterprise documents (4,869 pages) across 67 document schemas and 8 business domains. It jointly evaluates systems across the four dimensions producti