Jina OCR vs docTR
Pick Jina OCR to turn PDF pages into Markdown, HTML or LaTeX for research or non-commercial work. Pick docTR when you need word-level text with coordinates, or when you ship a commercial product, because it is Apache-2.0. The main split is output and licence: Jina OCR gives document structure under CC-BY-NC-4.0, while docTR gives words and lines under a permissive licence.
Jina OCRjina/jina-ocr | docTRmindee/doctr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 1.5 s / page | 0.9 s / page |
| Best for | Clean Markdown at low cost | Word-level OCR with positions |
| Inputs | PDF, images | PDF, images |
| Languages | About 100 languages | Mainly Latin scripts |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | ✓ Yes |
| 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 Jina OCR if
- You want page-to-Markdown, HTML or LaTeX output, which docTR does not produce.
- You want a published benchmark: 91.14 on OmniDocBench v1.6 (authors) and 83.4 on olmOCR-Bench, 7.4 points above DeepSeek-OCR.
- You can self-host or call Jina's hosted API, with only about 570M active parameters and 2.57 pages per second on an A100.
Choose docTR if
- You need Apache-2.0 for commercial use, while Jina OCR needs a separate license from Jina/Elastic.
- You need a nested page, block, line and word structure with coordinates as JSON, or bounding boxes for invoices and receipts feeding custom extraction.
- You want to choose detection and recognition architectures, use straight or rotated boxes, and run on CPU or NVIDIA GPU from ready Docker images.
Key differences
| Jina OCR | docTR | |
|---|---|---|
| Licence | CC-BY-NC-4.0; commercial use needs a license from Jina/Elastic | Apache-2.0 |
| How you run it | Self-host the open weights or call Jina's hosted API | Python library on PyTorch, with Docker images; no hosted API listed |
| Output formats | Markdown, HTML, LaTeX | JSON, plain text, bounding boxes |
| Tables, formulas and Markdown | Page-to-Markdown model with LaTeX output | No table, formula or Markdown output; returns words and lines only |
| Hardware | GPU recommended (A100 and L4 benchmarks); falls back to CPU in Transformers | CPU or NVIDIA GPU; GPU Docker images need CUDA 12.2 or newer on the host |
| Benchmarks | OmniDocBench v1.6 91.14, olmOCR-Bench 83.4 (authors; stricter prompt than the default) | Not published |
| Known failure | Hosted API returns HTTP 503 on cold start; retry after 30 to 60 seconds | Cropped images give poorer results (issue #2007); ONNX export of MASTER truncates decoding steps (issue #2224) |
Sources
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina OCR detailsdocTR
Open-source OCR library from Mindee with separate text detection and recognition models.
docTR 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": "jina/jina-ocr","fallbacks": ["mindee/doctr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesdocTR alternativesBest OCR models in 2026