Mathpix Convert vs PaddleOCR-VL
Pick Mathpix Convert for math-heavy papers and textbooks where you need LaTeX or DOCX out of whole PDFs, and you accept a paid hosted API. Pick PaddleOCR-VL for multilingual scans, tables and charts that must stay on your own GPU under an Apache-2.0 licence. The main split is a proprietary per-page API versus a free self-hosted model you must run and test yourself.
Mathpix Convertmathpix/convert | PaddleOCR-VLpaddle/paddleocr-vl | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 7 credits / page | 6 credits / page |
| Per 1,000 pages | $7.00 | $6.00 |
| Typical speed | 1.5 s / page | 1.3 s / page |
| Best for | Math and scientific papers | CJK documents |
| Inputs | PDF, images | PDF, images |
| Languages | 30+ languages | 109 languages |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | ✓ Yes | ✓ Yes |
| Forms | No | No |
| Multilingual | No | ✓ 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 Mathpix Convert if
- You need LaTeX, DOCX or HTML straight from whole PDFs, not just Markdown and JSON.
- Your documents are math-heavy papers or textbooks, or contain handwritten equations and chemistry.
- You want a hosted API with published per-page pricing and no GPU to run.
Choose PaddleOCR-VL if
- You must self-host on-prem, and the Apache-2.0 licence has no per-page fee.
- You process scans in many scripts. It supports 109 languages, including Cyrillic, Arabic, Devanagari and Thai, in one model.
- You parse reports with tables and charts on modest GPUs, since the model is under 1B parameters.
Key differences
| Mathpix Convert | PaddleOCR-VL | |
|---|---|---|
| How you run it | Hosted API | Self-host via PaddleOCR, vLLM or transformers; NVIDIA GPU recommended, CPU is slow |
| Licence | Proprietary | Apache-2.0 |
| Price | PDF $0.005 per page (0-1M), $0.0035 above 1M; images $0.002 each; one-time $19.99 API key activation fee | Free; you pay for compute |
| Output formats | LaTeX, Markdown, DOCX, HTML, JSON | Markdown, JSON |
| Benchmarks | Not published on general OCR leaderboards | 96.33 on OmniDocBench v1.6 (authors); 80.0 on olmOCR-Bench (independent) |
| Languages | Not published | 109 languages in one model |
| Known failure | Images with more than 12 text rows may be billed at the PDF page rate | Version 1.6 regresses Russian and Kazakh Cyrillic on born-digital PDFs, hangs on blurry images, and ships with use_cache false, which makes decoding about 11x slower |
Sources
- Mathpix billing docs
- Mathpix OCR API
- PaddleOCR-VL-1.6 model card
- olmOCR-Bench table
- Issue #18088 Cyrillic regression
- Issue #18453 use_cache slowdown
What each one is
Mathpix Convert
Specialist for math, science papers and handwritten formulas. Outputs LaTeX.
Mathpix Convert detailsPaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-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": "mathpix/convert","fallbacks": ["paddle/paddleocr-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Mathpix Convert alternativesPaddleOCR-VL alternativesBest OCR models in 2026