Marker vs docTR
Pick Marker when you need PDF to Markdown, HTML or JSON with equations and RAG chunks, and your company is under $5M or you buy a license. Pick docTR when you need word-level text with coordinates under a plain Apache-2.0 licence, and you will build any layout or extraction logic yourself. The main split is output: Marker returns structured documents, docTR returns words and lines only.
Markerdatalab/marker | docTRmindee/doctr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 14 credits / page | 1 credit / page |
| Per 1,000 pages | $14.00 | $1.00 |
| Typical speed | 2.5 s / page | 0.9 s / page |
| Best for | Books and research papers | Word-level OCR with positions |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | 90+ languages | Mainly Latin scripts |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | ✓ Yes |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
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
- You want Markdown, HTML, JSON block trees or RAG chunks straight from a PDF.
- You convert books and papers with equations and want LaTeX output.
- You need a CPU option: fast mode runs on CPU with about 18 MB of ONNX weights at 23.6 pages per second.
Choose docTR if
- You need the Apache-2.0 licence on both code and use, with no revenue cap on commercial use.
- You want word-level OCR with coordinates, as nested page, block, line and word JSON.
- You are a Python ML team that wants to pick detection and recognition architectures, use straight or rotated boxes, and train your own models.
Key differences
| Marker | docTR | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M license, free for research, personal use and startups under $5M. | Apache-2.0 |
| How you run it | Self-hosted on GPU, CPU or Apple Silicon (MPS), or Datalab's hosted API. | Self-hosted Python library on PyTorch, on CPU or NVIDIA GPU, with Docker images. GPU images need CUDA 12.2 or newer on the host. |
| Price | Free self-hosted within the weight license. Hosted API includes about 5,000 free pages per month, paid plans on Datalab's pricing page. | Free, self-hosted. You pay for compute. |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images. | JSON, plain text, bounding boxes. No table, formula or Markdown output. |
| Benchmarks | 83.7 on olmOCR-bench and 85.9 on OmniParseBench in vllm mode; CPU fast mode scores 66.6 on olmOCR-bench. All numbers are Datalab's own. | Not published |
| Tables | Not published | Not supported. It returns words and lines only. |
| Known failure | CPU fast mode drops to 66.6 on olmOCR-bench; commercial use above $5M needs a paid license. | Split-crop merging can misjoin readings across repeated characters (issue #2229). Cropped images give poorer results (issue #2007). ONNX export of MASTER truncates decoding steps (issue #2224). |
Sources
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsdocTR
Open-source OCR library from Mindee with separate text detection and recognition models.
docTR detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "datalab/marker","fallbacks": ["mindee/doctr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesdocTR alternativesBest OCR models in 2026