GPT vs Nougat
Pick GPT for handwritten notes, custom JSON extraction and document question answering through a hosted API, where you can pay per token. Pick Nougat for English arXiv-style papers with heavy math that must become LaTeX, but only if the use is non-commercial, because its weights are CC-BY-NC and it is self-hosted.
GPTopenai/gpt | Nougatmeta/nougat | |
|---|---|---|
| Type | Vision model | Open source |
| Price | 11 credits / page | 2 credits / page |
| Per 1,000 pages | $11.00 | $2.00 |
| Typical speed | 2.4 s / page | 3.0 s / page |
| Best for | Messy scans, custom instructions | arXiv-style papers |
| Inputs | PDF, images | PDF, images |
| Languages | 50+ languages | Mainly English |
| Max pages per request | 100 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | 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 GPT if
- It reads handwriting well: an HN user reported about 5% errors on messy cursive versus about 30% for Mistral OCR.
- It can extract data into custom JSON and answer questions about a document, not only transcribe it.
- It has no licence barrier for products: it is a hosted API that takes files up to 50 MB per request and returns Markdown, JSON or plain text.
Choose Nougat if
- It is built for scientific papers: trained on arXiv and PubMed Central, it outputs Mathpix-style Markdown with LaTeX for equations and tables.
- It costs nothing in fees for non-commercial use: you self-host it and pay only for compute.
- It ships small and base checkpoints with a simple CLI and API, and the code is MIT licensed.
Key differences
| GPT | Nougat | |
|---|---|---|
| How you run it | Hosted API from OpenAI | Self-hosted Python package |
| Licence | Proprietary | Code MIT; weights CC-BY-NC, so no commercial use |
| Pricing | Token-based per model; page images add input tokens, so cost per page is not fixed | Free for non-commercial use; you pay for compute |
| Output formats | Markdown, JSON, plain text | Markdown (Mathpix-style .mmd) and LaTeX |
| Best document types | Handwritten notes, custom JSON extraction, document question answering | English arXiv-style papers and math-heavy PDFs |
| Languages | Not published | Works best on English; Chinese, Russian and Japanese do not work |
| Known failure | No bounding boxes or confidence scores in the standard response | Repetition loops on 1.5% of test pages, more on out-of-domain documents; looping pages can be dropped as MISSING_PAGE |
Sources
- OpenAI file inputs docs
- HN Mistral OCR thread
- OpenAI API pricing
- Nougat GitHub README
- Nougat paper
- Issue #11 repetitions
What each one is
Try 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": "openai/gpt","fallbacks": ["meta/nougat"],"file_url": "https://example.com/document.pdf"}