Chandra vs Granite Docling
Pick Chandra for multilingual scans, forms and math-heavy pages where you want the highest published olmOCR-bench score (85.8) and a hosted API option. Pick Granite Docling for English PDFs with tables, code and equations that you want to convert cheaply on a CPU or small GPU inside Docling, under a plain Apache-2.0 licence. The main split is language coverage and licence: Chandra covers 43 languages but its weights carry a revenue limit, while Granite Docling is English-first and fully free to use.
Chandradatalab/chandra | 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 | 2.0 s / page | 1.5 s / page |
| Best for | Complex layouts and handwriting | Small-model document conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 40+ languages | English, experimental CJK and Arabic |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | ✓ Yes | 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 Chandra if
- It scores 85.8 on olmOCR-bench and averages 77.8% across 43 languages, so it fits multilingual scanned documents.
- It is available as a hosted Datalab API at $1, $3 or $7.50 per 1,000 pages, so you can skip running GPUs yourself.
- It returns Markdown, HTML and JSON with layout chunks, bounding boxes and labels per block, which suits PDF to Markdown for RAG.
Choose Granite Docling if
- It is Apache-2.0 with no revenue cap, while Chandra's weights need a commercial licence for companies above $2M funding or revenue.
- At 258M parameters it is built for low-cost CPU or small GPU conversion, while Chandra is run on a single NVIDIA GPU such as an H100, A100, L40S or RTX 4090.
- It scores 0.97 TEDS structure on FinTabNet tables and F1 0.968 on equations and 0.988 on code, so it suits English PDFs with tables, code and equations.
Key differences
| Chandra | Granite Docling | |
|---|---|---|
| Licence | Code Apache-2.0; weights modified OpenRAIL-M, free for research, personal use and companies under $2M funding/revenue | Apache-2.0 |
| How you run it | Self-host with vLLM or Hugging Face, or use the hosted Datalab API | Self-host only, typically inside Docling |
| Price | Self-hosted free within licence terms; API $1, $3 or $7.50 per 1,000 pages; free tier of $20/month | Free, you pay for compute |
| Hardware | Single NVIDIA GPU via vLLM; benchmarks on one H100 80GB | 258M parameters; suited to CPU or small GPU |
| Languages | 77.8% average over 43 languages; weak on Urdu (43.2%) and Sanskrit (51.1%) | English primary; Japanese, Arabic and Chinese marked experimental |
| Output formats | Markdown, HTML, JSON, bounding boxes | DocTags, Markdown, HTML, LaTeX, bounding boxes |
| Known failure | Token-loop hallucinations on some Russian/Kazakh scans (issue #51); Old Scans category only 51.1 on olmOCR-bench | Repetitive loops on page headings and invoices; may output only exclamation marks on GPUs without bfloat16 such as T4 unless run in float32 |
Sources
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
- Granite-Docling model card
- HF discussion #24 repetition
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra 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": "datalab/chandra","fallbacks": ["ibm/granite-docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Chandra alternativesGranite Docling alternativesBest OCR models in 2026