OCRRouter

SmolDocling vs Chunkr

Pick Chunkr if you need RAG chunking of PDFs, slides and Word files, or a hosted API with a self-host option. Pick SmolDocling only for English page-conversion experiments on small hardware, since it is a preview model that its authors have replaced with Granite-Docling.

SmolDoclinghuggingface/smoldoclingChunkrchunkr/chunkr
TypeOpen sourceOCR API
Price1 credit / page14 credits / page
Per 1,000 pages$1.00$14.00
Typical speed1.0 s / page3.0 s / page
Best forVery cheap page conversionLayout-aware chunks for RAG
InputsPDF, imagesPDF, images, DOCX
LanguagesMainly EnglishMultilingual
Max pages per request5001,000
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
FormsNoNo
MultilingualNo✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose SmolDocling if

  • It is Apache-2.0 licensed per its model summary and free to run, so you pay only for compute.
  • At 256M parameters it runs on modest hardware, and it takes about 0.35 seconds per page on an A100 with vLLM.
  • One model outputs text, tables, code, formulas and charts with location tags, in DocTags, Markdown, HTML, JSON and LaTeX.

Choose Chunkr if

  • It accepts PDF, PPT, Word and images, and returns HTML, Markdown, JSON and OCR with bounding boxes.
  • It offers both a cloud API and an open-source self-host version, with Docker options for GPU, CPU-only and Mac ARM.
  • It does layout analysis and semantic chunking, which suits RAG pipelines over PDFs and slides.

Key differences

SmolDoclingChunkr
DeploymentSelf-host onlyCloud API or self-host via Docker
LicenceApache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0)AGPL-3.0 or commercial licence; closed-source products need the commercial one
PriceFree, you pay for computeHosted credit-based plans (prices not published); open-source version is free but needs compute and an LLM endpoint
Output formatsDocTags, Markdown, HTML, JSON, LaTeX, bounding boxesHTML, Markdown, JSON, bounding boxes
BenchmarksTEDS table structure 0.82 on FinTabNet, full-page OCR F1 0.80, OCRBench 338 (author-reported)Not published
LanguagesEnglish onlyNot published
Known failureOutput can loop on some images, with endless token repetition or missing closing tags, which makes generation very slowOpen-source version uses community models that are less accurate than the cloud's, and does not support Excel

Sources

What each one is

SmolDocling

Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.

SmolDocling details

Chunkr

Document parsing API that segments and chunks pages for RAG. Self-hostable version available.

Chunkr details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "huggingface/smoldocling","fallbacks": ["chunkr/chunkr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

SmolDocling alternativesChunkr alternativesBest OCR models in 2026