Jina OCR vs Qwen-VL
Pick Jina OCR for non-commercial page-to-Markdown work, where it scores 83.4 on olmOCR-Bench and uses only about 570M active parameters. Pick Qwen-VL when you need an Apache-2.0 licence for a commercial product, or one model that also handles photos, screenshots and reasoning. Jina's licence is the main limit: CC BY-NC 4.0 blocks commercial use without a separate deal.
Jina OCRjina/jina-ocr | Qwen-VLqwen/qwen-vl | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.5 s / page | 2.0 s / page |
| Best for | Clean Markdown at low cost | Flexible extraction with prompts |
| Inputs | PDF, images | PDF, images |
| Languages | About 100 languages | 30+ languages |
| Max pages per request | 200 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| 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
- It is built for page-to-Markdown and also outputs HTML and LaTeX.
- It scores 91.14 on OmniDocBench v1.6 (authors' figure) and 83.4 on olmOCR-Bench, 7.4 points above DeepSeek-OCR.
- Only about 570M parameters are active, and it runs at 2.57 pages per second on an A100.
Choose Qwen-VL if
- Apache-2.0 allows commercial use with no separate license deal.
- It comes in sizes from 2B to 235B-A22B MoE, and outputs bounding boxes, JSON, HTML and Markdown.
- It reads text in 32 languages and handles photos and screenshots, with low light, blur and tilted text covered per Qwen.
Key differences
| Jina OCR | Qwen-VL | |
|---|---|---|
| Licence | CC-BY-NC-4.0; commercial use needs a license from Jina/Elastic | Apache-2.0 (Qwen3-VL repository) |
| How you run it | Self-host or Jina's hosted API | Self-host open weights or Alibaba Cloud hosted API |
| Model size | 3B MoE, about 570M active | 2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE |
| Output formats | Markdown, HTML, LaTeX | Plain text, HTML, bounding boxes, JSON, Markdown |
| olmOCR-Bench | 83.4 overall (authors) | 64.6 for Qwen 3 VL 8B (independent, Datalab table) |
| Hardware | GPU recommended (A100, L4); CPU fallback in Transformers | FP8 checkpoints need H100-class GPU and CUDA 12+ |
| Known failure | Hosted API returns HTTP 503 on cold start; retry after 30 to 60 seconds | Can loop repeating the same text, worse with greedy decoding and on low-quality scans; Qwen2.5-VL-32B GGUF often loops on scanned documents |
Sources
- jina-ocr-v1 model card
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
- llama.cpp issue #14663
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina OCR detailsQwen-VL
Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.
Qwen-VL 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": ["qwen/qwen-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesQwen-VL alternativesBest OCR models in 2026