Marker vs GOT-OCR 2.0
Pick Marker when you need to turn PDFs into Markdown, HTML or JSON, especially books and papers with equations, and when you want RAG chunks or a hosted API option. Pick GOT-OCR 2.0 when you need a self-hosted model for formulas, sheet music, chemical structures or region-specific OCR on English and Chinese documents. The main split is the job: Marker is a document converter with several output formats, GOT-OCR 2.0 is a single 580M-parameter model with no published benchmark scores.
Markerdatalab/marker | GOT-OCR 2.0stepfun/got-ocr-2 | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 14 credits / page | 1 credit / page |
| Per 1,000 pages | $14.00 | $1.00 |
| Typical speed | 2.5 s / page | 1.5 s / page |
| Best for | Books and research papers | Formatted text, formulas and charts |
| Inputs | PDF, images, DOCX | images |
| Languages | 90+ languages | Chinese, English |
| Max pages per request | 1,000 | 100 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| 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 want PDF to Markdown, HTML, JSON block trees or RAG chunks, and Marker outputs all of them.
- You need to run on a CPU or Apple Silicon: fast mode uses about 18 MB of ONNX weights and runs at 23.6 pages per second on CPU.
- You want a hosted option: Datalab sells an API with about 5,000 free pages per month.
Choose GOT-OCR 2.0 if
- You need to read formulas, sheet music or chemical structures, and it outputs LaTeX, TikZ, SMILES and kern.
- You want to OCR one region of a page, guided by box coordinates or color.
- You want a single 580M-parameter model that is in Hugging Face Transformers with batched inference and multi-page OCR.
Key differences
| Marker | GOT-OCR 2.0 | |
|---|---|---|
| How you run it | Self-host on GPU, CPU or Apple Silicon, or use Datalab's hosted API | Self-host only; CUDA 11.8 and PyTorch 2.0.1 per the README, with a llama.cpp port for CPU users |
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M license, free for research, personal use and startups under $5M | Code Apache-2.0; data CC BY-NC 4.0 |
| Price | Free self-hosted within the weight license, you pay for compute; hosted API has about 5,000 free pages per month | Free self-hosted, you pay for compute |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images | Plain text, Markdown, LaTeX, TikZ, SMILES, kern |
| Benchmarks | 83.7 on olmOCR-bench in vllm mode and 85.9 on OmniParseBench, both reported by Datalab; CPU fast mode scores 66.6 on olmOCR-bench | Not published |
| Languages | Not published | English and Chinese documents |
| Known failure | Weight license blocks commercial use for companies above $5M without a paid license | Issue #226 reports output looping characters midway; issue #235 reports wrong text on images with no text; issue #317 reports format mode is much worse than plain OCR mode |
Sources
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsGOT-OCR 2.0
Unified OCR model that reads plain text, tables, formulas, sheet music and charts.
GOT-OCR 2.0 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": ["stepfun/got-ocr-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesGOT-OCR 2.0 alternativesBest OCR models in 2026