PaddleOCR-VL vs Qwen-VL
Pick PaddleOCR-VL for self-hosted document parsing: scanned multilingual pages, academic papers with formulas, and tables and charts in reports. It is under 1B parameters, supports 109 languages and scores higher on OCR benchmarks. Pick Qwen-VL when OCR is one task among several, such as screenshots, photos with text, or questions about a document, or when you want a hosted API.
PaddleOCR-VLpaddle/paddleocr-vl | Qwen-VLqwen/qwen-vl | |
|---|---|---|
| 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 | 2.0 s / page |
| Best for | CJK documents | Flexible extraction with prompts |
| Inputs | PDF, images | PDF, images |
| Languages | 109 languages | 30+ languages |
| Max pages per request | 500 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | ✓ Yes | 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 PaddleOCR-VL if
- It is a dedicated document parser at 0.9B parameters, so it runs on much smaller GPUs than Qwen-VL's larger sizes.
- It supports 109 languages, including Cyrillic, Arabic, Devanagari and Thai, against 32 for Qwen-VL.
- It scores 80.0 on olmOCR-Bench, while Qwen 3 VL 8B scores 64.6 on the same benchmark in a separate independent table.
Choose Qwen-VL if
- It is a general vision-language model, so one model can do OCR plus reasoning or visual question answering.
- It comes in sizes from 2B to 235B-A22B MoE, all Apache-2.0, and Alibaba Cloud also serves it as a hosted API.
- It outputs bounding boxes and a Qwen HTML document format as well as plain text, JSON and Markdown, and Qwen says it handles low light, blur and tilted text.
Key differences
| PaddleOCR-VL | Qwen-VL | |
|---|---|---|
| Licence | Apache-2.0 | Apache-2.0 (Qwen3-VL repository) |
| How you run it | Self-host only, via PaddleOCR, vLLM or transformers | Self-host the open weights, or use the hosted API on Alibaba Cloud |
| Model size | 0.9B | 2B, 4B, 8B, 32B dense; 30B-A3B and 235B-A22B MoE |
| Output formats | Markdown, JSON | Plain text, HTML, bounding boxes, JSON, Markdown |
| Languages | 109 languages in one model | OCR in 32 languages |
| olmOCR-Bench | 80.0 (PaddleOCR-VL) | 64.6 (Qwen 3 VL 8B) |
| Known failure | Version 1.6 regresses Russian and Kazakh OCR on born-digital PDFs, can hang on blurry images, and ships with use_cache false, which makes decoding about 11x slower | Can loop and repeat the same text, mostly on low-quality images and worse with greedy decoding; GGUF builds often get stuck on scanned documents |
Sources
- Hugging Face model card
- olmOCR-Bench table
- Issue #18088 Cyrillic regression
- Qwen3-VL GitHub
- Issue #1611 infinite repetition
- Chandra README benchmark table
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL 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": "paddle/paddleocr-vl","fallbacks": ["qwen/qwen-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesQwen-VL alternativesBest OCR models in 2026