ExtractBench Pro (LlamaIndex)
by LlamaIndex
What is ExtractBench Pro (LlamaIndex)? ExtractBench Pro (LlamaIndex) is an image & vision eval made by LlamaIndex. The harder companion to ExtractBench, released October 2026: 100 synthetic documents (2,622 pages, 86 document types, 144,141 expected values) where many values cannot be copied from the page. They must be read off charts or drawings, counted under the document's own rules, or decoded with its own jargon. Same evaluator as ExtractBench; the headline metric is the share of documents extracted perfectly.
Editor note
The harder companion to ExtractBench, released October 2026: 100 synthetic documents (2,622 pages, 86 document types, 144,141 expected values) where many values cannot be copied from the page. They must be read off charts or drawings, counted under the document's own rules, or decoded with its own jargon. Same evaluator as ExtractBench; the headline metric is the share of documents extracted perfectly.
- What it tests
- Extraction systems on long-tail documents with charts, diagrams and house rules.
- Original source
- www.extractbench.ai/pro
- Last updated
- Dataset updated 2026-10-20
- License
- Apache-2.0
Reviews
No reviews yet.
Reviews from practitioners who sign in with GitHub open soon.