SmolDocling vs Typhoon OCR
Pick Typhoon OCR if your documents are Thai or Thai-English, such as government forms, financial reports or books. Pick SmolDocling only for English experiments on tiny hardware, and move to Granite-Docling for anything in production. Typhoon OCR is the one with a language focus and a hosted API. SmolDocling is a 256M preview model that its authors have superseded.
SmolDoclinghuggingface/smoldocling | Typhoon OCRtyphoon/typhoon-ocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.0 s / page | 1.8 s / page |
| Best for | Very cheap page conversion | Thai documents |
| Inputs | PDF, images | PDF, images |
| Languages | Mainly English | Thai, English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| 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 SmolDocling if
- It has 256M parameters, so it runs on modest hardware, and it takes about 0.35 seconds per page on an A100 with vLLM.
- One model handles text, tables, code, formulas and charts, and it returns location tags and bounding boxes.
- It outputs DocTags for Docling, plus Markdown, HTML, JSON and LaTeX, which suits English page conversion on a low budget.
Choose Typhoon OCR if
- It is built for Thai and English. V1.5 scored BLEU 0.870 on Thai government forms, ahead of GPT-5 and Gemini 2.5 Pro.
- You can self-host the weights or call the Typhoon API as typhoon-ocr-v2, so you can start hosted and move to your own servers later.
- It returns tables as HTML with merged cells, equations as LaTeX and marked checkboxes. V2 2B scores 84.7 on olmOCR-bench.
Key differences
| SmolDocling | Typhoon OCR | |
|---|---|---|
| Languages | English only | Thai and English; no other language was evaluated |
| Deployment | Self-host only | Self-host, or the Typhoon API as typhoon-ocr-v2 |
| Model size | 256M | 2B (V2 and V1.5), plus a 0.8B V2 variant |
| Licence | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 | Apache-2.0, subject to OpenTyphoon Terms and Conditions |
| Output formats | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes | Markdown, HTML, LaTeX |
| Benchmarks | Table TEDS 0.82 on FinTabNet at 150dpi, full-page OCR F1 0.80, OCRBench 338 | olmOCR-bench 84.7, OmniDocBench v1.6 90.97, ThaiOCRBench 59.9 (all V2 2B) |
| Known failure | Output can loop endlessly on some images (Docling issue #1515), and the paper lists missing closing tags | V2 can loop until the token limit, mostly on dense multi-column pages. Key-information extraction is weakest, at 21.0 on ThaiOCRBench |
Sources
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
- Typhoon OCR V2 2B model card
- Typhoon OCR technical report
What each one is
SmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsTyphoon OCR
Open vision model tuned for Thai and English documents, with Markdown output.
Typhoon OCR 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": "huggingface/smoldocling","fallbacks": ["typhoon/typhoon-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
SmolDocling alternativesTyphoon OCR alternativesBest OCR models in 2026