Datalab Releases Open Document Extraction Benchmark

Datalab has released OmniExtractBench, an open benchmark for structured document extraction that tests how accurately systems fill a JSON schema from a PDF. It pools 620 documents from four existing benchmarks and uses one deterministic scorer that explains each decision, aiming to replace vendor leaderboards Datalab calls hard to compare or audit. The scorer installs from PyPI as omni-extract-bench.