Unlimited OCR vs Docling
Pick Unlimited OCR when you need to parse long scanned reports or books in one pass on a CUDA GPU and want text with bounding boxes. Pick Docling when your inputs are mixed files (PDF, DOCX, PPTX, XLSX, HTML) and you need structured JSON on any machine, including CPU-only ones. The main split is scope: Unlimited OCR is a single GPU model for long documents, while Docling is a conversion library that depends on the OCR engine you plug in.
Unlimited OCRbaidu/unlimited-ocr | Doclingibm/docling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 1.2 s / page | 2.2 s / page |
| Best for | Books and very long documents | Structured export for RAG |
| Inputs | PDF, images | PDF, images, DOCX |
| Languages | 30+ languages | Latin scripts, CJK |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| 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 Unlimited OCR if
- It parses 40+ pages in a single forward pass with roughly constant memory, so long reports and books go in as one request.
- It emits block types and bounding boxes alongside text, which suits layout-aware extraction.
- Baidu reports 77.1 for Unlimited OCR 2 on MarathonDocBench against 33.7 for the next model.
Choose Docling if
- It reads PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRL through one API.
- Its lossless JSON export keeps layout, reading order and table structure for downstream tools.
- It runs on macOS, Linux and Windows, on x86_64 and arm64, and works on CPU.
Key differences
| Unlimited OCR | Docling | |
|---|---|---|
| What it is | 3B Mixture-of-Experts document parser from Baidu, derived from DeepSeek-OCR | Python library from IBM Research that converts documents and plugs in OCR engines and VLMs |
| Licence | MIT | MIT (individual models have their own licences) |
| How you run it | Self-host with vLLM, SGLang or transformers; also hosted in Baidu Cloud's OCR product | Local Python library; free, you pay for compute |
| Hardware | NVIDIA GPU with CUDA 12.9 or 13.0; version 2 under SGLang runs out of memory on 12 GB GPUs; no CPU-only | CPU works, GPU speeds up model stages |
| Output formats | Markdown, plain text, bounding boxes | Markdown, HTML, JSON, DocTags, plain text |
| Benchmarks | ParseBench mean 46.17 (independent); MPDocBench 84.6 and MarathonDocBench 77.1 for version 2 (authors) | Not published |
| Known failure | Text looping and hallucinated text (issues #55, #58); text formatting score of 0.97 on ParseBench, so bold and italic are mostly lost; context capped at 32,768 tokens | Version 2.143.1 drops words and numbers with full-page Tesseract OCR (issue #5254); neighbouring columns can merge in reading order (issue #5132) |
Sources
- Hugging Face model card
- MarkTechPost on Unlimited OCR 2
- Issue #55 repetition
- Docling GitHub README
- Supported formats
- Issue #5254 OCR regression
What each one is
Unlimited 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": "baidu/unlimited-ocr","fallbacks": ["ibm/docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Unlimited OCR alternativesDocling alternativesBest OCR models in 2026