OCRRouter

Marker vs SmolDocling

Pick Marker for self-hosted PDF to Markdown, especially books and papers with equations, and for RAG chunking. It has published benchmarks and a hosted API, and its fast mode runs on CPU. Pick SmolDocling only for small English experiments on modest hardware, since it is a preview release and the authors point users to Granite-Docling.

Markerdatalab/markerSmolDoclinghuggingface/smoldocling
TypeOCR APIOpen source
Price14 credits / page1 credit / page
Per 1,000 pages$14.00$1.00
Typical speed2.5 s / page1.0 s / page
Best forBooks and research papersVery cheap page conversion
InputsPDF, images, DOCXPDF, images
Languages90+ languagesMainly English
Max pages per request1,000500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
FormsNoNo
Multilingual✓ YesNo
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 Marker if

  • Marker reports 83.7 on olmOCR-bench in vllm mode, and it outputs Markdown, HTML, JSON block trees and RAG chunks.
  • Fast mode runs on CPU at 23.6 pages per second with about 18 MB of ONNX weights, though the score drops to 66.6 on olmOCR-bench.
  • Datalab sells a hosted API with about 5,000 free pages per month, so you can skip running it yourself.

Choose SmolDocling if

  • It has 256M parameters, small enough for modest hardware, and runs at about 0.35 seconds per page on an A100 with vLLM.
  • The Apache-2.0 licence (per the model summary) has no revenue cap, while Marker's weights block free commercial use for companies above $5M.
  • One model handles text, tables, code, formulas and charts, and it outputs DocTags with bounding boxes.

Key differences

MarkerSmolDocling
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and startups under $5MApache-2.0 per the model summary (the metadata tag says CDLA-Permissive-2.0)
How you run itSelf-host on GPU, CPU or Apple Silicon, or use the Datalab hosted APISelf-host only
PriceFree self-hosted within the weight licence; hosted API has about 5,000 free pages per month, then paid plansFree self-hosted; you pay for compute
Output formatsMarkdown, HTML, JSON, chunks, LaTeX, imagesDocTags, Markdown, HTML, JSON, LaTeX, bounding boxes
BenchmarksolmOCR-bench 83.7 and OmniParseBench 85.9 in vllm mode, all reported by DatalabTable TEDS 0.82 on FinTabNet, full-page OCR F1 0.80, OCRBench 338, all reported by the authors
LanguagesNot publishedEnglish only
Known failureCPU fast mode drops to 66.6 on olmOCR-benchEndless token repetition and missing closing tags; output loops on some images (Docling issue #1515)
StatusMaintained, with a hosted optionPreview release, superseded by Granite-Docling

Sources

What each one is

Marker

Hosted version of the open-source Marker and Surya models. Strong on books and papers.

Marker details

SmolDocling

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

SmolDocling 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": "datalab/marker","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Marker alternativesSmolDocling alternativesBest OCR models in 2026