TrOCR vs Typhoon OCR
Pick TrOCR if you already have cropped single-line images, such as English handwriting or receipt lines, and want an MIT-licensed recognizer to fine-tune. Pick Typhoon OCR if you need to read whole Thai or English pages with tables and equations. The main split is scope: TrOCR reads one text line at a time, while Typhoon OCR outputs structured Markdown, HTML or LaTeX for full documents.
TrOCRmicrosoft/trocr | 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 | 1.2 s / page | 1.8 s / page |
| Best for | Handwritten text lines | Thai documents |
| Inputs | images | PDF, images |
| Languages | English | Thai, English |
| Max pages per request | 100 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | No | No |
| Bounding boxes | No | 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 TrOCR if
- You have English handwritten lines: the large model reports 2.89 character error rate on the IAM handwriting set.
- You want to fine-tune a recognizer on your own line images, starting from separate handwritten, printed (SROIE) and scene text checkpoints.
- You want the simplest licence and setup: MIT, and it loads in a few lines with Hugging Face transformers.
Choose Typhoon OCR if
- Your documents are Thai or Thai-English mixed, such as government forms, financial reports or books. V1.5 beat GPT-5 and Gemini 2.5 Pro on Thai government forms (BLEU 0.870).
- You need page-level output: tables as HTML with merged cells, equations as LaTeX, and checkboxes marked.
- You want a choice of self-hosting or the Typhoon API (typhoon-ocr-v2), with a 2B model that scores 84.7 on olmOCR-bench.
Key differences
| TrOCR | Typhoon OCR | |
|---|---|---|
| Input it reads | Single text-line images only; you need a separate line detector | Full document pages with layout |
| Output formats | Plain text | Markdown, HTML and LaTeX |
| Languages | English-focused checkpoints; other languages need fine-tuning | Thai and English; no other languages evaluated |
| Licence | MIT (microsoft/unilm repository) | Apache-2.0, subject to OpenTyphoon Terms and Conditions |
| How you run it | Locally through Hugging Face transformers | Self-hosted weights or the Typhoon API (typhoon-ocr-v2) |
| Model size | Base 334M, large 558M parameters | 2B (V2 and V1.5); 0.8B V2 variant |
| Known failure | Slow versus CRNN recognizers on CPU; no layout, table or page output | V2 can loop until the token limit on dense multi-column pages; weakest at key-information extraction (21.0 on ThaiOCRBench) |
Sources
- TrOCR paper (arXiv 2109.10282)
- trocr-large-handwritten model card
- Transformers TrOCR docs
- Typhoon OCR V2 2B model card
- Typhoon OCR technical report
What each one is
TrOCR
Transformer text recognition model. Reads single text lines, including handwriting.
TrOCR 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": "microsoft/trocr","fallbacks": ["typhoon/typhoon-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
TrOCR alternativesTyphoon OCR alternativesBest OCR models in 2026