olmOCR vs Donut
Pick olmOCR to turn full PDF pages into Markdown for RAG or LLM training, including papers with equations and pages with handwriting. Pick Donut when you have a narrow, fixed-layout document type such as receipts or forms and can fine-tune a model to extract fields as JSON. The main split is that olmOCR works out of the box on whole pages, while Donut is a base checkpoint that must be trained before use.
olmOCRallenai/olmocr | Donutnaver/donut | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.5 s / page | 1.0 s / page |
| Best for | Bulk PDF to text for datasets | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | Mainly English | English, Chinese, Japanese, Korean |
| Max pages per request | 1,000 | 100 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | No | ✓ Yes |
| Multilingual | No | 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 olmOCR if
- You need full-page PDF to Markdown output without training anything, and it scores 82.4 on olmOCR-Bench.
- Your documents include equations, tables or handwriting, and you want headers and footers stripped in natural reading order.
- You process large volumes: self-hosting costs under $200 per million pages per Ai2, and hosted olmOCR-2-7B on DeepInfra costs $0.09 input / $0.19 output per million tokens.
Choose Donut if
- You extract fields from receipts or forms with a fixed layout and can fine-tune a model for them.
- You want a permissive MIT licence, and the model loads in Hugging Face Transformers with no external OCR step.
- You classify documents, or you want JSON output straight from the image. One HN practitioner reports about 200 fine-tuning documents can reach human-level accuracy on narrow data.
Key differences
| olmOCR | Donut | |
|---|---|---|
| Out-of-the-box use | Converts PDFs and images to Markdown as shipped | Base checkpoint is pretrain-only and must be fine-tuned before use |
| Licence | Apache-2.0 (toolkit) | MIT |
| How you run it | Self-host with vLLM on a GPU, or call third-party hosted endpoints | Self-host only; you pay for compute |
| Hardware | NVIDIA GPU with at least 12 GB VRAM and 30 GB free disk. Out of memory reported on a 12 GB GPU (issue #424) | Not published |
| Output formats | Markdown, JSON | JSON, plain text |
| Benchmark | 82.4 on olmOCR-Bench (v0.4.0), above Marker (76.1), MinerU 2.5.4 (75.2) and Mistral OCR (72.0) | Not published |
| Known failure | Can get stuck on the last few pages (issue #433). Image bounding box coordinates are unreliable (issue #437) | Misses tiny text in large images because of input resolution limits. Transformers v5 dropped image-to-text pipeline support, so use direct loading or v4 |
Sources
- olmOCR GitHub README
- Issue #424 OOM
- Issue #433 stuck
- donut-base model card
- Donut ECCV paper
- Hacker News thread
What each one is
olmOCR
Open-source vision model from Ai2 (often searched as olm ocr) that turns PDFs into clean, reading-order text. Latest release: olmOCR 2.
olmOCR 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.
Use one, fall back to the other
{"model": "allenai/olmocr","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
olmOCR alternativesDonut alternativesBest OCR models in 2026