Chandra vs Donut
Pick Chandra for full-page conversion of scanned PDFs to Markdown, HTML or JSON, especially multilingual ones, with no training step. Pick Donut only if you have a narrow, fixed-layout document type such as receipts or forms and will fine-tune it. The main split: Chandra works out of the box, Donut is a base model you must train first.
Chandradatalab/chandra | Donutnaver/donut | |
|---|---|---|
| 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 | 1.0 s / page |
| Best for | Complex layouts and handwriting | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | 40+ languages | English, Chinese, Japanese, Korean |
| Max pages per request | 500 | 100 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | ✓ Yes | ✓ Yes |
| 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
- You need whole pages as Markdown, HTML or JSON, with bounding boxes and labels per block, for RAG pipelines.
- You work with multilingual scans: the authors report 77.8% average across 43 languages and 85.8 on olmOCR-bench.
- You want a hosted option too: the Datalab API costs $1, $3 or $7.50 per 1,000 pages, with $20/month free usage.
Choose Donut if
- You want an MIT licence with no revenue or funding cap on commercial self-hosting.
- You have a narrow, fixed-layout document type, such as receipts or forms, and will fine-tune for field extraction or classification.
- You want to skip a separate OCR step: it maps the document image straight to structured JSON, and a practitioner reports about 200 fine-tuning documents can reach human-level accuracy on narrow data.
Key differences
| Chandra | Donut | |
|---|---|---|
| Out-of-the-box use | Converts pages to Markdown, HTML or JSON with layout as released | Base checkpoint is pretrain-only and must be fine-tuned before use |
| Licence | Code Apache-2.0; weights modified OpenRAIL-M, free for research, personal use and companies under $2M funding/revenue | MIT |
| How you run it | Self-host with vLLM or Hugging Face, or use the hosted Datalab API | Self-host only, via Hugging Face Transformers |
| Price | Self-hosted: free within licence terms. API: $1, $3 or $7.50 per 1,000 pages | Free, self-hosted (you pay for compute) |
| Output formats | Markdown, HTML, JSON, bounding boxes | JSON, plain text |
| Published benchmarks | olmOCR-bench 85.8 (Chandra 2); 77.8% across 43 languages | Not published |
| Known failure | Token-loop hallucinations reported on some Russian/Kazakh scans; weak on Urdu (43.2%) and old scans | Misses tiny text in large images because of input resolution limits; Transformers v5 dropped image-to-text pipeline support, so load it directly or use v4 |
Sources
- Chandra GitHub README
- Datalab pricing
- GitHub issue #51 token loops
- donut-base model card
- Donut ECCV paper
- Hacker News thread
What each one is
Chandra
Datalab's OCR model that keeps layout and handles forms, tables and handwriting.
Chandra detailsDonut
OCR-free document understanding model. Fine-tuned for receipts, forms and document Q&A.
Donut detailsTry both on your own file
Run the file once with each model and compare the output.
PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?
Use one, fall back to the other
{"model": "datalab/chandra","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Chandra vs GOT-OCR 2.0Chandra vs NougatSurya vs DonutolmOCR vs DonutPaddleOCR (PP-OCR) vs DonutChandra vs Nanonets-OCRChandra vs Typhoon OCRDonut vs GOT-OCR 2.0
Chandra alternativesDonut alternativesBest OCR models in 2026