PaddleOCR-VL vs Nanonets-OCR
Pick PaddleOCR-VL for self-hosted parsing of multilingual scans, academic papers with formulas, and tables and charts, especially on modest GPUs. Pick Nanonets-OCR for printed business documents that need signatures, watermarks, checkboxes or flowcharts tagged in the output. The main reason is accuracy and licensing: PaddleOCR-VL is Apache-2.0 and scores 80.0 on olmOCR-Bench, while Nanonets-OCR scores 69.5 and its model card states no licence.
PaddleOCR-VLpaddle/paddleocr-vl | Nanonets-OCRnanonets/nanonets-ocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 6 credits / page | 2 credits / page |
| Per 1,000 pages | $6.00 | $2.00 |
| Typical speed | 1.3 s / page | 1.8 s / page |
| Best for | CJK documents | Markdown with semantic tags |
| Inputs | PDF, images | PDF, images |
| Languages | 109 languages | Multilingual |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | ✓ Yes | No |
| Forms | No | ✓ Yes |
| 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 PaddleOCR-VL if
- It is Apache-2.0 licensed, so the terms for commercial self-hosting are clear.
- It scores 80.0 on olmOCR-Bench (an independent listing) versus 69.5 for Nanonets-OCR2-3B (reported by its authors), and supports 109 languages including Cyrillic, Arabic, Devanagari and Thai.
- At under 1B parameters it runs on much smaller GPUs than a 3B model, and it outputs Markdown and JSON for pages with text, tables, formulas and charts.
Choose Nanonets-OCR if
- It tags signatures, watermarks and image descriptions, and converts checkboxes to unicode symbols, which suits forms and business documents.
- It extracts flowcharts and org charts as Mermaid code, and outputs HTML and LaTeX as well as Markdown, with inline and display equations told apart.
- It reports 89.43 on DocVQA and 78.56 on ChartQA for the 3B model, and PaddleOCR-VL publishes no such figures.
Key differences
| PaddleOCR-VL | Nanonets-OCR | |
|---|---|---|
| Licence | Apache-2.0 | Not stated on the Nanonets-OCR2-3B model card; base model is Qwen2.5-VL-3B |
| Size | 0.9B parameters | 3B parameters (Hugging Face lists about 4B) |
| Output formats | Markdown, JSON | Markdown, HTML, LaTeX, Mermaid |
| olmOCR-Bench overall | 80.0 (independent) | 69.5 (reported by authors) |
| Languages | 109 languages in one model | Not published |
| Prompting | Not published | Trained on a fixed prompt; hallucination with other prompts is expected |
| Known failures | Version 1.6 has a Russian and Kazakh Cyrillic regression, use_cache false makes decoding about 11x slower, and inference hangs on blurry images | Not trained on handwriting and can hallucinate; one handwriting test saw generation loops with CER up to 2100% and 69.1s per image on an L4 |
Sources
- Hugging Face model card
- olmOCR-Bench table
- Issue #18453 use_cache slowdown
- Nanonets-OCR2-3B model card
- Nanonets-OCR-s announcement
- HF discussion fixed prompt
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL detailsNanonets-OCR
Open image-to-Markdown model that tags tables, equations, signatures, watermarks and checkboxes.
Nanonets-OCR 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": "paddle/paddleocr-vl","fallbacks": ["nanonets/nanonets-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesNanonets-OCR alternativesBest OCR models in 2026