TrOCR vs Nougat
Pick Nougat for English scientific papers where you need Markdown with LaTeX math and tables, as long as your product is non-commercial. Pick TrOCR for reading single text-line images, such as English handwriting or receipt lines, under an MIT licence. The main split is scope and licence: Nougat parses whole PDF pages but its weights are CC-BY-NC, while TrOCR reads one line at a time and needs a separate line detector.
TrOCRmicrosoft/trocr | Nougatmeta/nougat | |
|---|---|---|
| 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 | 3.0 s / page |
| Best for | Handwritten text lines | arXiv-style papers |
| Inputs | images | PDF, images |
| Languages | English | Mainly English |
| Max pages per request | 100 | 200 |
| Tables | No | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | No | ✓ Yes |
| Bounding boxes | No | No |
| 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 TrOCR if
- You need a commercial-friendly licence, since the TrOCR code is MIT licensed and Nougat's weights are CC-BY-NC.
- You are reading English handwritten lines, and the large model reports a 2.89 character error rate on IAM.
- You want to fine-tune a recognizer on your own line images, or use the separate checkpoints for handwritten (IAM), printed (SROIE) and scene text.
Choose Nougat if
- You are converting English arXiv-style papers or math-heavy PDFs to Markdown with LaTeX equations and tables.
- You need page-level output from a PDF, which TrOCR does not give you.
- Your use is non-commercial and you want a simple CLI and API with small and base checkpoints.
Key differences
| TrOCR | Nougat | |
|---|---|---|
| Input unit | Single text-line images only; you need a separate line detector | Scientific PDF pages |
| Output | Plain text | Markdown (Mathpix-style .mmd) with LaTeX for math and tables |
| Licence | MIT (microsoft/unilm repository) | Code MIT; weights CC-BY-NC, so no commercial use |
| Price and hosting | Free, self-hosted; you pay for compute | Free for non-commercial use, self-hosted; you pay for compute |
| Model size | Base 334M, large 558M parameters | Small and base checkpoints; parameter counts not published |
| Languages | Released checkpoints are English-focused; other languages need fine-tuning | Works best on English; Chinese, Russian and Japanese do not work |
| Known failure | No layout, table or page-level output; slow versus CRNN recognizers on CPU | Repetition loops on 1.5% of test pages, more on out-of-domain documents; looping pages can be dropped as MISSING_PAGE |
Sources
- TrOCR paper (arXiv 2109.10282)
- trocr-large-handwritten model card
- Nougat GitHub README
- Nougat paper
- Issue #11 repetitions
What each one is
TrOCR
Transformer text recognition model. Reads single text lines, including handwriting.
TrOCR detailsTry both on your own file
Run the file once with each model and compare the output.
PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?
Use one, fall back to the other
{"model": "microsoft/trocr","fallbacks": ["meta/nougat"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
TrOCR vs DonutTrOCR vs MonkeyOCRolmOCR vs NougatKraken vs NougatMathpix Convert vs NougatTrOCR vs Adobe Acrobat OCRMarker vs NougatTrOCR vs Google Cloud Vision OCR
TrOCR alternativesNougat alternativesBest OCR models in 2026