PaddleOCR-VL vs PaddleOCR (PP-OCR)
Pick PaddleOCR-VL when you need full pages parsed into Markdown or JSON, with tables, formulas and charts, across 109 languages, and you have a GPU. Pick PP-OCR when you need fast text lines and boxes on CPU, edge devices or in the browser, mainly for printed Chinese and English. The main split is output: PP-OCR returns lines and boxes only, while PaddleOCR-VL returns structured documents.
PaddleOCR-VLpaddle/paddleocr-vl | PaddleOCR (PP-OCR)paddle/paddleocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 6 credits / page | 1 credit / page |
| Per 1,000 pages | $6.00 | $1.00 |
| Typical speed | 1.3 s / page | 0.7 s / page |
| Best for | CJK documents | Fast printed text, Chinese and English |
| Inputs | PDF, images | PDF, images |
| Languages | 109 languages | 80+ languages |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | ✓ Yes | ✓ Yes |
| 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 need whole pages converted to Markdown or JSON, including tables, formulas and charts.
- You work with scanned documents in many scripts, since one model covers 109 languages including Cyrillic, Arabic, Devanagari and Thai.
- You can run a GPU and want a document parser under 1B parameters, much smaller than 7B document VLMs like olmOCR.
Choose PaddleOCR (PP-OCR) if
- You only need text lines and bounding boxes, for example from receipts, signs and screenshots.
- You must run on CPU, mobile or edge hardware, or in the browser through the PaddleOCR.js SDK.
- You want a 34.5M parameter model (PP-OCRv6 medium) that reports 0.13s on an A100 and a 5.2x CPU speedup with OpenVINO.
Key differences
| PaddleOCR-VL | PaddleOCR (PP-OCR) | |
|---|---|---|
| Licence and price | Apache-2.0, free, self-hosted (you pay for compute) | Apache-2.0, free, self-hosted (you pay for compute) |
| How you run it | Self-hosted via PaddleOCR, vLLM or transformers. The transformers path only does element-level recognition, not full page parsing. | Library: paddleocr Python package, C++ or the PaddleOCR.js browser SDK |
| Model size | 0.9B parameters | 34.5M parameters (PP-OCRv6 medium) |
| Hardware | NVIDIA GPU recommended. CPU is possible but slow. No minimum VRAM stated. | CPU, NVIDIA GPU, Kunlunxin XPU and other accelerators |
| Output | Markdown and JSON, covering text, tables, formulas and charts | Plain text, bounding boxes and JSON. Tables and formulas need the separate PP-Structure or VL pipelines. |
| Languages | 109 languages, including Cyrillic, Arabic, Devanagari and Thai | 50 languages in one PP-OCRv6 model: Chinese, English, Japanese and 46 Latin-script languages |
| Benchmark | OmniDocBench v1.6 overall 96.33 (PaddleOCR-VL-1.6, by the authors). olmOCR-Bench overall 80.0 (independent). | Not published |
| Known failure | Version 1.6 hangs on blurry images, decodes about 11x slower because use_cache ships as false, and regresses Russian and Kazakh OCR on born-digital PDFs versus 1.5. | PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower. |
Sources
- PaddleOCR-VL-1.6 model card
- PaddleOCR GitHub README
- olmOCR-Bench table
- Issue #18453 use_cache slowdown
- PP-OCRv6 on Hugging Face
- RapidOCR issue #743 denormal weights
What each one is
PaddleOCR-VL
Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.
PaddleOCR-VL detailsPaddleOCR (PP-OCR)
Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.
PaddleOCR (PP-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": ["paddle/paddleocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR-VL alternativesPaddleOCR (PP-OCR) alternativesBest OCR models in 2026