metadata
license: mit
size_categories:
- 1K<n<10K
A benchmark of real documents paired with ground-truth JSON, designed to score end-to-end accuracy rather than raw character error rate. That distinction matters: an OCR pass can be 99% correct at the character level and still get the invoice total wrong.
We use it for: scoring candidate OCR models on the metric that actually pays the bills - regression testing before a model swap.
Attribution
This is an unmodified fork of
getomni-ai/ocr-benchmark, created by the Qwen team. All weights, files and behaviour are identical to upstream — we rehost it so our experiments stay reproducible and version-pinned. The original license and all credit remain with the Qwen team. If you want the canonical dataset, please use the original.