OCRmyPDF vs Qwen-VL
Pick OCRmyPDF to make scanned PDF archives searchable as PDF/A on CPU, with no GPU and no layout output. Pick Qwen-VL when you need text from photos or screenshots, or Markdown, HTML, JSON and bounding boxes from a document. The main split is a Tesseract text layer on the original PDF versus a general vision-language model that returns structured output but can loop on low-quality scans.
OCRmyPDFocrmypdf/ocrmypdf | 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 | 1.0 s / page | 2.0 s / page |
| Best for | Searchable PDFs from scans | Flexible extraction with prompts |
| Inputs | PDF, images | PDF, images |
| Languages | 100+ languages (via Tesseract) | 30+ languages |
| Max pages per request | 1,000 | 200 |
| Tables | No | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | No | ✓ 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 OCRmyPDF if
- It outputs validated searchable PDF/A by default, which suits archiving and compliance workflows.
- It runs on CPU only and spreads work across all cores, so you need no GPU.
- It inserts the OCR layer losslessly where possible without re-encoding other content in the PDF.
Choose Qwen-VL if
- It returns Markdown, HTML, JSON and bounding boxes, not just plain text or a text layer.
- It handles photos and screenshots with text, and per Qwen copes with low light, blur and tilted text.
- It comes in sizes from 2B to 235B-A22B MoE under Apache-2.0, and Alibaba Cloud also serves it as a hosted API.
Key differences
| OCRmyPDF | Qwen-VL | |
|---|---|---|
| Licence | MPL-2.0 | Apache-2.0 (Qwen3-VL repository) |
| How you run it | Local command-line tool, CPU only, uses all cores | Self-hosted open weights or Alibaba Cloud hosted API. FP8 checkpoints need an H100-class GPU and CUDA 12+ |
| Price | Free, self-hosted (you pay for compute) | Free, self-hosted (you pay for compute). Hosted pricing varies by model and is not published here |
| Output formats | Searchable PDF/A, plain text | Plain text, HTML, bounding boxes, JSON, Markdown |
| Languages | 100+ languages via Tesseract language packs | OCR in 32 languages |
| Benchmark | Not published | Qwen 3 VL 8B scores 64.6 on olmOCR-bench (independent, Datalab table), below dedicated OCR models |
| Known failure | Output size can double or quadruple with --deskew --clean-final (#948). --skip-text with default PDF/A output removes form fields (#1790). It hangs on PDFs whose forms share one Resources dictionary (#1763) | Can loop repeating the same text, worse with greedy decoding and on low-quality images (#1611, #241). Qwen2.5-VL-32B GGUF often loops on scanned documents (llama.cpp #14663) |
Sources
- OCRmyPDF GitHub README
- Issue #948 file size
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
- llama.cpp issue #14663
What each one is
OCRmyPDF
Command-line tool that adds a searchable text layer to scanned PDFs, using Tesseract.
OCRmyPDF 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": "ocrmypdf/ocrmypdf","fallbacks": ["qwen/qwen-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
OCRmyPDF alternativesQwen-VL alternativesBest OCR models in 2026