Surya vs RolmOCR
Pick Surya when you need layout, reading order, bounding boxes or HTML tables, and when you want to run on a laptop or CPU. Pick RolmOCR when you only need plain text from scanned pages, including rotated ones, and you have a GPU. Surya has a licence limit above $5M revenue or funding. RolmOCR is plain Apache-2.0.
Suryadatalab/surya | RolmOCRreducto/rolmocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Layout and line-level OCR | Fast open PDF to text |
| Inputs | PDF, images | PDF, images |
| Languages | 90+ languages | Not stated by the authors |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | No | No |
| Bounding boxes | ✓ Yes | No |
| 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 Surya if
- It returns layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes from one model.
- It runs on CPU and Apple Silicon through llama.cpp, with peak RAM of about 4-6 GB on an M4 Max.
- At 689M parameters it scores 83.7% on olmOCR-bench and reaches 4.52 pages/s at 192 DPI on one H100 with vLLM.
Choose RolmOCR if
- The weights are Apache-2.0 and usable commercially, with no revenue or funding cap.
- About 15% of its training data was rotated, so it suits off-angle and skewed scans.
- Needl.ai counted 7 hallucination issues for it versus 11 for olmocr-2-7b, though its overall score was lower (0.663 vs 0.770).
Key differences
| Surya | RolmOCR | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding/revenue. | Apache-2.0 |
| Model size | 689M parameters | 8B (Qwen2.5-VL-7B base) |
| How you run it | Lightweight client against a vLLM or llama.cpp server. NVIDIA GPU via vLLM, or CPU/Apple Silicon via llama.cpp. | Self-hosted, for example with vLLM, on one GPU. |
| Output formats | JSON, HTML, LaTeX, bounding boxes, plain text | Plain text only |
| Bounding boxes | Per-block and line-level boxes for 90+ languages | Cannot output layout bounding boxes |
| Tables | HTML tables, but Surya 2.5 output lacks row/column geometry. Multilingual tables score only 58.1. | Not suited to structured table output. Reducto shows it omitting a table subtitle. |
| Benchmark | 83.7% on olmOCR-bench (reported by the authors) | 0.663 overall on the Needl.ai PDF extraction benchmark (independent) |
Sources
- Surya GitHub README
- Datalab pricing and license FAQ
- Hugging Face model card
- Needl.ai benchmark
- Reducto announcement
What each one is
Surya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya detailsRolmOCR
Reducto's open PDF-to-text model, built on Qwen2.5-VL-7B with the olmOCR recipe. Needs no PDF metadata and handles rotated pages.
RolmOCR 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/surya","fallbacks": ["reducto/rolmocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Surya alternativesRolmOCR alternativesBest OCR models in 2026