OCRRouter

Qwen-VL vs Nougat

Pick Qwen-VL for general OCR on photos, screenshots and mixed documents, especially if you need a commercial licence and more than English. Pick Nougat only for English scientific papers where you want LaTeX math, and only for non-commercial work. The licence and language coverage decide most cases: Qwen-VL is Apache-2.0 with OCR in 32 languages, while Nougat weights are CC-BY-NC and fail on Chinese, Russian and Japanese.

Qwen-VLqwen/qwen-vlNougatmeta/nougat
TypeOpen sourceOpen source
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed2.0 s / page3.0 s / page
Best forFlexible extraction with promptsarXiv-style papers
InputsPDF, imagesPDF, images
Languages30+ languagesMainly English
Max pages per request200200
Tables✓ Yes✓ Yes
Handwriting✓ YesNo
Math✓ Yes✓ Yes
Bounding boxes✓ YesNo
ConfidenceNoNo
FormsNoNo
Multilingual✓ YesNo
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose Qwen-VL if

  • Its code and weights are Apache-2.0 across all sizes from 2B to 235B-A22B, so you can ship it in a commercial product.
  • It reads OCR in 32 languages and handles photos, screenshots, blur, low light and tilted text per Qwen.
  • It outputs plain text, HTML, Markdown, JSON and bounding boxes, and can also answer questions about the document (VQA).

Choose Nougat if

  • It is built for academic papers and was trained on arXiv and PubMed Central documents.
  • It writes LaTeX for equations and tables in Mathpix-style Markdown, which suits math-heavy PDFs.
  • It is a small self-hosted package with small and base checkpoints and a simple CLI and API.

Key differences

Qwen-VLNougat
LicenceApache-2.0Code MIT; weights CC-BY-NC, so no commercial use
How you run itSelf-host the open weights, or use the hosted API on Alibaba CloudSelf-host only, as a Python package
PriceFree self-hosted, you pay for compute. Hosted pricing varies by model and is not published hereFree for non-commercial use, you pay for compute
Output formatsPlain text, HTML, Markdown, JSON, bounding boxesMarkdown and LaTeX
LanguagesOCR in 32 languagesWorks best on English; Chinese, Russian and Japanese do not work
BenchmarkQwen 3 VL 8B scores 64.6 on olmOCR-bench, below dedicated OCR models (independent table)Not published
Known failureCan loop and repeat the same text, worse with greedy decoding and on low-quality scans; transformers 5.17 caused loops on 10-37% of OCR outputs versus 2-3% baselineRepetition loops on 1.5% of test pages, more on out-of-domain documents; the repetition check can drop pages as MISSING_PAGE

Sources

What each one is

Qwen-VL

Alibaba's open-weight vision language models (Qwen2.5-VL, Qwen3-VL). Strong general OCR and grounding.

Qwen-VL details

Nougat

Meta's model that converts academic PDFs to Markdown with LaTeX math.

Nougat details

Try 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": "qwen/qwen-vl","fallbacks": ["meta/nougat"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Qwen-VL alternativesNougat alternativesBest OCR models in 2026