Marker vs PaddleOCR-VL
Pick Marker if you want PDF to Markdown on your own machine, including CPU or Apple Silicon, and you fit its weight licence. Pick PaddleOCR-VL if you need Apache-2.0 for all of it, many languages, or tables, formulas and charts from scans on a modest GPU. Licence and hardware are the main split: Marker's weights block free commercial use above $5M, while PaddleOCR-VL needs a GPU to run at a useful speed.
Markerdatalab/marker | PaddleOCR-VLpaddle/paddleocr-vl | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 14 credits / page | 6 credits / page |
| Per 1,000 pages | $14.00 | $6.00 |
| Typical speed | 2.5 s / page | 1.3 s / page |
| Best for | Books and research papers | CJK documents |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | 90+ languages | 109 languages |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | ✓ 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 Marker if
- You need to run on CPU, Apple Silicon or a GPU, and fast mode runs on CPU at 23.6 pages per second with about 18 MB of ONNX weights.
- You want more output formats: Markdown, HTML, JSON block trees, RAG chunks, LaTeX and images.
- You want a hosted option too, since Datalab's API includes about 5,000 free pages per month.
Choose PaddleOCR-VL if
- You want Apache-2.0 with no revenue cap, which matters for companies above $5M.
- You parse multilingual documents, since one model covers 109 languages including Cyrillic, Arabic, Devanagari and Thai.
- You have a modest GPU, because the model is under 1B parameters and reads text, tables, formulas and charts on full pages.
Key differences
| Marker | PaddleOCR-VL | |
|---|---|---|
| Licence | Code Apache-2.0. Weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and startups under $5M. | Apache-2.0 |
| How you run it | Self-host on GPU, CPU or MPS, or use Datalab's hosted API. vllm mode needs an NVIDIA GPU. | Self-host only, through PaddleOCR, vLLM or transformers. NVIDIA GPU recommended; CPU is possible but slow. |
| Price | Free self-hosted within the weight licence. Hosted API has about 5,000 free pages per month, then paid plans. | Free self-hosted. You pay for compute. |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images | Markdown, JSON |
| Benchmark | 83.7 on olmOCR-bench in vllm mode, reported by Datalab. 66.6 in CPU fast mode. | 80.0 on olmOCR-Bench, from an independent table. Authors report 96.33 on OmniDocBench v1.6 for version 1.6. |
| Languages | Not published | 109 languages in one model |
| Known failure | CPU fast mode scores lower on olmOCR-bench. All README benchmark numbers are Datalab's own. | Version 1.6 regresses Russian and Kazakh Cyrillic on born-digital PDFs, hangs on blurry images, and ships use_cache false, which makes decoding about 11x slower. |
Sources
- Marker GitHub README
- Datalab pricing
- Hugging Face model card
- olmOCR-Bench table
- Issue #18088 Cyrillic regression
- Issue #18453 use_cache slowdown
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker 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": "datalab/marker","fallbacks": ["paddle/paddleocr-vl"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesPaddleOCR-VL alternativesBest OCR models in 2026