olmOCR vs SmolDocling
Pick olmOCR when you need accurate PDF to Markdown at volume on a GPU, including papers with equations and pages with handwriting. Pick SmolDocling only to experiment with a 256M model on modest hardware, and only for English pages. The main reason is size and maturity: olmOCR is a 7B model that scores 82.4 on olmOCR-Bench, while SmolDocling is a 256M preview that its authors say is superseded by Granite-Docling.
olmOCRallenai/olmocr | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.5 s / page | 1.0 s / page |
| Best for | Bulk PDF to text for datasets | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | Mainly English | Mainly English |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | 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 olmOCR if
- It scores 82.4 on olmOCR-Bench, above Marker (76.1), MinerU 2.5.4 (75.2) and Mistral OCR (72.0) in the same table.
- It handles equations, tables and handwriting, and strips headers and footers in natural reading order.
- Self-hosting costs under $200 per million pages per Ai2, and a hosted olmOCR-2-7B endpoint on DeepInfra costs $0.09 input / $0.19 output per million tokens.
Choose SmolDocling if
- At 256M parameters it is small enough for modest hardware, and it runs at about 0.35 seconds per page on an A100 with vLLM.
- It outputs DocTags, Markdown, HTML, JSON, LaTeX and bounding boxes, so you get location tags that olmOCR does not reliably give.
- One model covers text, tables (OTSL), code, formulas and charts, and it is free to self-host apart from compute.
Key differences
| olmOCR | SmolDocling | |
|---|---|---|
| Model size | 7B parameters | 256M parameters |
| Status | Current toolkit with an olmOCR-2-7B model | Preview release, superseded by Granite-Docling |
| How you run it | Self-host with vLLM on an NVIDIA GPU with at least 12 GB VRAM, or use hosted endpoints | Self-host only |
| Licence | Apache-2.0 (toolkit) | Apache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0) |
| Output formats | Markdown, JSON | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Languages | Not published | English only |
| Known failure | Out of memory on a 12 GB GPU (issue #424), stuck on the last few pages (issue #433), unreliable image bounding boxes (issue #437) | Endless token repetition and missing closing tags; output loops on some images and generation becomes very slow (Docling issue #1515) |
Sources
- olmOCR GitHub README
- Issue #424 OOM
- Issue #437 bbox
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
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 detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling 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": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
olmOCR alternativesSmolDocling alternativesBest OCR models in 2026