olmOCR vs Surya
Pick Surya if you need layout, reading order and bounding boxes, or want to run OCR on a laptop or CPU. It is a 689M model that scores 83.7% on olmOCR-bench, but its licence needs a commercial deal above $5M in revenue or funding. Pick olmOCR for large-scale PDF to Markdown with equations, tables and handwriting, under an Apache-2.0 toolkit, if you have an NVIDIA GPU with at least 12 GB VRAM.
olmOCRallenai/olmocr | Suryadatalab/surya | |
|---|---|---|
| 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 | 1.5 s / page |
| Best for | Bulk PDF to text for datasets | Layout and line-level OCR |
| Inputs | PDF, images | PDF, images |
| Languages | Mainly English | 90+ languages |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | ✓ 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 olmOCR if
- You need handwriting, equations and tables converted to Markdown, and olmOCR is listed for mixed printed and handwritten pages and academic papers.
- You are processing PDFs at scale for LLM training or RAG, where self-hosting costs under $200 per million pages, per Ai2.
- You want the Apache-2.0 toolkit with no revenue cap, or a hosted option: DeepInfra charges $0.09 input and $0.19 output per million tokens for olmOCR-2-7B.
Choose Surya if
- You need layout labels, reading order and per-block bounding boxes from one model, in 90+ languages.
- You have no big GPU: Surya runs on CPU and Apple Silicon through llama.cpp, with peak memory of about 4.3-5.9 GB on an M4 Max.
- You want a small model (689M parameters) that is fast on a GPU: 4.52 pages/s at 192 DPI on one H100 with vLLM.
Key differences
| olmOCR | Surya | |
|---|---|---|
| Licence | Apache-2.0 (toolkit) | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding/revenue. |
| Model size and hardware | 7B. NVIDIA GPU with at least 12 GB VRAM and 30 GB free disk. No CPU-only use. | 689M. NVIDIA GPU via vLLM, or CPU/Apple Silicon via llama.cpp. |
| How you run it | Self-host with vLLM, or call third-party hosted endpoints. | Self-host only. A lightweight client talks to a vLLM or llama.cpp server you run yourself. |
| Price | Self-hosted: under $200 per million pages, per Ai2. Hosted via DeepInfra: $0.09 input / $0.19 output per million tokens. | Free to self-host (you pay for compute). Commercial licence needed above $5M revenue or funding. |
| Output formats | Markdown, JSON | JSON, HTML, LaTeX, bounding boxes, plain text |
| olmOCR-bench score | 82.4 (olmOCR-Bench overall, v0.4.0) | 83.7% (Surya 2.5), as reported by its authors |
| Known failures | Out of memory on a 12 GB GPU (issue #424), stuck on the last few pages (issue #433), unreliable image bounding box coordinates (issue #437). | Surya 2.5 table output lacks row/column geometry. Multilingual tables score 58.1. Only 21 of 91 languages reach a 90% pass rate. Not for photos or natural scenes. |
Sources
What each one is
olmOCR
Open-source vision model from Ai2 (often searched as olm ocr) that turns PDFs into clean, reading-order text. Latest release: olmOCR 2.
olmOCR detailsSurya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya 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": "allenai/olmocr","fallbacks": ["datalab/surya"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
olmOCR alternativesSurya alternativesBest OCR models in 2026