PaddleOCR-VL vs Chunkr
Pick PaddleOCR-VL if you want a small, Apache-2.0 model you run yourself for multilingual scans, academic papers with formulas, and tables or charts in reports. Pick Chunkr if you need a parsing service that chunks PDFs and slides for RAG, with a hosted API or a self-host option. The main split is that PaddleOCR-VL is a 0.9B recognition model with a permissive licence, while Chunkr is a layout and chunking pipeline under AGPL-3.0 or a commercial licence.
PaddleOCR-VLpaddle/paddleocr-vl | Chunkrchunkr/chunkr | |
|---|---|---|
| Type | Open source | OCR API |
| Price | 6 credits / page | 14 credits / page |
| Per 1,000 pages | $6.00 | $14.00 |
| Typical speed | 1.3 s / page | 3.0 s / page |
| Best for | CJK documents | Layout-aware chunks for RAG |
| Inputs | PDF, images | PDF, images, DOCX |
| Languages | 109 languages | Multilingual |
| Max pages per request | 500 | 1,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| 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
- You want an Apache-2.0 licence, so you can use it in closed-source products without buying a commercial licence.
- You process documents in many languages: it supports 109, including Cyrillic, Arabic, Devanagari and Thai scripts, in one model.
- You have a modest GPU: at under 1B parameters it needs far less hardware than 7B document VLMs, and it needs no external LLM endpoint.
Choose Chunkr if
- You are building RAG and want semantic chunking of PDFs, PPT, Word files and images, with bounding boxes on the OCR output.
- You want the choice of a hosted cloud API or a Docker self-host with GPU, CPU-only and Mac ARM options.
- You already have an OpenAI-compatible LLM endpoint and want HTML, Markdown or JSON output from a single service.
Key differences
| PaddleOCR-VL | Chunkr | |
|---|---|---|
| Licence | Apache-2.0 | AGPL-3.0 or a commercial licence. Closed-source use needs the commercial licence. |
| How you run it | Self-host only, via PaddleOCR, vLLM or transformers. NVIDIA GPU recommended; CPU works but is slow. | Hosted cloud API or self-host via Docker (GPU, CPU-only, Mac ARM). Self-host needs an external LLM endpoint. |
| Price | Free. You pay for compute. | Hosted: credit-based plans, prices not published here. Open-source version is free, you pay for compute and an LLM endpoint. |
| Output formats | Markdown and JSON | HTML, Markdown, JSON and bounding boxes |
| Input types | Not published | PDF, PPT, Word and images. No Excel in the open-source version. |
| Benchmarks | OmniDocBench v1.6 overall 96.33 (version 1.6, reported by the authors). olmOCR-Bench overall 80.0 (independent). | Not published |
| Known issues | Version 1.6 regresses Russian and Kazakh OCR on born-digital PDFs, hangs on blurry images, and can miss table content. Its default use_cache false setting makes decoding about 11x slower. | The open-source version uses community models that are less accurate than the cloud's in-house models. |
Sources
- Hugging Face model card
- PaddleOCR-VL-1.6 model card
- olmOCR-Bench table
- Issue #18453 use_cache slowdown
- Chunkr GitHub README
- Chunkr site
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL detailsChunkr
Document parsing API that segments and chunks pages for RAG. Self-hostable version available.
Chunkr 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": ["chunkr/chunkr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesChunkr alternativesBest OCR models in 2026