Unlimited OCR vs Granite Docling
Pick Unlimited OCR for long reports and books that you want parsed in one request, on an NVIDIA GPU with enough memory. Pick Granite Docling for English PDFs with tables, code and equations, or when you need to run on a CPU or a small GPU. The main split is size and scope: 3B parameters and 40+ pages per pass versus 258M parameters inside the Docling library.
Unlimited OCRbaidu/unlimited-ocr | Granite Doclingibm/granite-docling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.2 s / page | 1.5 s / page |
| Best for | Books and very long documents | Small-model document conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 30+ languages | English, experimental CJK and Arabic |
| 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 through in one request.
- It emits block types and bounding boxes with the text, which suits layout-aware extraction and bulk PDF to text for RAG.
- A hosted endpoint is available in Baidu Cloud's OCR product if you do not want to run the GPU yourself.
Choose Granite Docling if
- At 258M parameters it suits low-cost CPU or small GPU conversion, where Unlimited OCR 2 runs out of memory on 12 GB GPUs.
- It reports table structure TEDS of 0.97 on FinTabNet, code F1 of 0.988 and equation F1 of 0.968, so it fits English PDFs with tables, code and equations.
- It outputs DocTags, Markdown, HTML, LaTeX and bounding boxes, and it is built to work inside Docling pipelines.
Key differences
| Unlimited OCR | Granite Docling | |
|---|---|---|
| Size | 3B total, about 500M active per token; Unlimited OCR 2 is 3.34B | 258M |
| Licence | MIT | Apache-2.0 |
| How you run it | Self-host with vLLM, SGLang or transformers, or use the hosted endpoint in Baidu Cloud | Self-host only, typically inside Docling |
| Hardware | 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 | Low-cost CPU or small GPU; on GPUs without bfloat16 such as T4 it may output only exclamation marks unless run in float32 |
| Output formats | Markdown, plain text, bounding boxes | DocTags, Markdown, HTML, LaTeX, bounding boxes |
| Multi-page input | 40+ pages in one pass; context capped at 32,768 tokens, and multi-page input only works in base mode | Not published |
| Known failure | Repetition loops (issue #55), hallucinated text (issue #58), and most bold and italic styling is lost | Repetitive loops on page headings and invoices; English primary, with Japanese, Arabic and Chinese marked experimental |
Sources
- Hugging Face model card (Unlimited OCR)
- GitHub repo (Unlimited OCR)
- Issue #116 OOM on 12 GB
- Granite-Docling model card
- HF discussion #24 repetition
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 detailsGranite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling 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/granite-docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Unlimited OCR alternativesGranite Docling alternativesBest OCR models in 2026