Marker vs Unlimited OCR
Pick Marker for self-hosted PDF to Markdown on books and papers with equations, or when you need to run on CPU or Apple Silicon, as long as your company is under $5M or you buy a license. Pick Unlimited OCR for long reports and books parsed in one request on an NVIDIA GPU, or when you need MIT-licensed commercial use. The main split is licence and hardware: Marker's model weights restrict larger companies and Unlimited OCR needs CUDA.
Markerdatalab/marker | Unlimited OCRbaidu/unlimited-ocr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 14 credits / page | 2 credits / page |
| Per 1,000 pages | $14.00 | $2.00 |
| Typical speed | 2.5 s / page | 1.2 s / page |
| Best for | Books and research papers | Books and very long documents |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | 90+ languages | 30+ languages |
| Max pages per request | 1,000 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| 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 have no NVIDIA GPU: Marker runs on CPU or Apple Silicon (MPS), and its fast mode uses about 18 MB of ONNX weights at 23.6 pages per second.
- You need more output formats: Markdown, HTML, JSON block trees, RAG chunks, LaTeX and images.
- You work with books and papers with equations, and you want a score you can check against a benchmark: 83.7 on olmOCR-bench in vllm mode (Datalab's own number).
Choose Unlimited OCR if
- You are a company above $5M and want free commercial use: the licence is MIT, with no copyleft.
- You parse long reports and books in one request: it handles 40+ pages in a single forward pass with roughly constant memory.
- You need layout-aware extraction: it returns block types and bounding boxes alongside the text.
Key differences
| Marker | Unlimited OCR | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M license, free for research, personal use and startups under $5M | MIT |
| Hardware | GPU, CPU or Apple Silicon (MPS); vllm mode needs an NVIDIA GPU | NVIDIA GPU with CUDA 12.9 or 13.0; version 2 under SGLang runs out of memory on 12 GB GPUs; not for CPU-only machines |
| Hosted option | Datalab API with about 5,000 free pages per month, paid plans after that | Hosted endpoint in Baidu Cloud's OCR product |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images | Markdown, plain text, bounding boxes |
| Benchmark | olmOCR-bench 83.7 and OmniParseBench 85.9 in vllm mode, both reported by Datalab; CPU fast mode drops to 66.6 on olmOCR-bench | ParseBench mean 46.17 (independent); Unlimited OCR 2 scores 84.6 on MPDocBench and 77.1 on MarathonDocBench, reported by Baidu |
| Long documents | Not published | 40+ pages in one pass; context capped at 32,768 tokens, and multi-page input only works in base mode |
| Known failure | Weight licence blocks free commercial use above $5M | Text repetition and looping (issue #55), hallucinated text (issue #58), bbox coordinate loops on GB10 GPUs under SGLang (issue #114), and mostly lost bold and italic styling (ParseBench text formatting 0.97) |
Sources
- Marker GitHub README
- Datalab pricing
- Hugging Face model card
- GitHub repo
- MarkTechPost on Unlimited OCR 2
- Issue #116 OOM on 12 GB
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsUnlimited OCR
Open-source model from Baidu that reads hundreds of pages in one pass with flat memory use.
Unlimited OCR 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": ["baidu/unlimited-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesUnlimited OCR alternativesBest OCR models in 2026