Put the information in your documents to work
Extract text, tables, and fields from PDFs, scans, invoices, and office files. Prepare readable content for an AI app or structured data for your next workflow.
Start with the document you have
A scanned invoice needs its text recovered before your app can use it. OCR handles that step and can preserve useful structure, such as headings and tables. A PDF that already contains selectable text may only need text extraction.
If you need specific fields, such as an invoice date or total, look for structured extraction. You can request annotations from a supported OCR API or send the extracted text to a model, as the example in the OCR guide shows.
A vision model can help when the task involves interpreting an image, such as classifying a document or checking what appears on a page. Compare its output with OCR on your own files. OCR is billed per page on the route shown here; a separate model request is billed by token usage.
Explore the guides
Compare providers
Compare starting prices and the tasks each provider supports. Features and usage affect the total cost. For measured results, see the benchmarks and how we test.
| Provider | Best suited for | Starting price |
|---|---|---|
| Anthropic | Ask Claude questions about document text or images. | metered by Anthropic |
| Grok | Work with documents using supported vision and text models. | metered by Grok (xAI) |
| Hugging Face | Use supported models for document analysis. | metered by Hugging Face |
| LlamaParse | Parse documents into content your application can use. | $0.00625 per call |
| Mistral Document AI | Read scanned pages with OCR and structured annotations. | $0.004 per call |
| OpenAI | Extract or interpret information with supported GPT models. | metered by OpenAI |
Use it with NativePort
Use document processing and AI models with the same NativePort key. One balance covers the document extraction and any model calls you use to work with the results.
Before you start
Should I use OCR or a vision model?
Start with OCR when you need the document text and layout. Use a vision model when you need to ask questions about what is on the page. If you need both, extract the text first and pass it to a model for the next step.
How do I send a document?
The Mistral OCR example accepts an HTTPS URL. Put your file in storage that the service can fetch, then pass its link and the pages you want to process. The request body is limited to 16 KB, so the file itself is not uploaded in that request.
How is document processing billed?
Mistral OCR is billed per page, with a separate rate for requests that include structured annotations. The response reports how many pages were processed. If you send the output to a model afterwards, that model request is billed separately. Check the provider catalog for prices.