Extract Hosted: the PDF extraction API
Coming soonSend a document.Get tables, headings, figures, chunks, text back.
One request returns text in reading order, tables, headings, figures and chunks a model can use. Nothing to install or run.
The Extract engine,
behind one request.
What it does
Extract Hosted is a PDF extraction API. You send a PDF or a scan, and the response carries the text in reading order, the tables as cells, the headings, the figures and chunks that are ready for a model or a vector store.
How it works
Every request names a quality. fast returns text and Markdown without models. balanced adds the table model. auto picks the models each page needs, and max also writes a description of every figure. Long documents run as async jobs and call your webhook when they are done.
Who it is for
Teams that build search or agents on documents and would rather not run the engine themselves. If your documents cannot leave your own machines, the Extract SDK does the same work inside your own process.
When
Coming soon. The API is not open yet; the SDK is available today. Prices are prepaid credit packs, set out on the hosted pricing page.
Three lines.
One answer.
1export PDFLUENT_KEY="your-key"
Tables that stay tables.
And the rest of the page too.
Tables, cell by cell
Figures you can search
Scans, read
Read well.
On the public benchmark.
Quality, quality auto
- PDFluent 1.4.00.9282
- opendataloader [hybrid]0.907
- nutrient0.885
- docling0.882
- marker0.861
- unstructured [hi_res]0.841
- edgeparse0.837
- opendataloader0.831
- mineru0.831
- pymupdf4llm0.732
- unstructured0.686
- markitdown0.589
- liteparse0.576
PDFluent 1.4.0 with quality auto, page headers and footers kept. Other engines as published in the benchmark README on 2 November 2026. Footnote
* Overall score 0.9282 (SDK 1.4.0, quality="auto", keep_furniture=True, layout, table and formula models installed) on all 200 documents, with the benchmark’s own harness and evaluator, measured 11 November 2026 on the final macOS arm64 release wheel. Competitor figures are taken from the benchmark’s public README as read on 3 November 2026 (highest overall there: 0.907, opendataloader [hybrid]), not re-run by us. The leaderboard engines ran on their own setups while PDFluent was measured on macOS arm64, and table structure depends on the platform. PDFluent is not on that leaderboard. Other engines have submissions open that are not yet on the leaderboard. One run, no confidence interval. The PDFluent row keeps page headers and footers because the benchmark’s ground truth counts them as content. PyMuPDF and MuPDF are trademarks of Artifex Software, Inc. PDFluent is an independent product of Innovation Trigger B.V. and is not affiliated with or endorsed by Artifex.
Measured with the SDK. Hosted runs the same engine; a measurement of the hosted API itself is coming.
Buy a pack once.
Use it page by page.
You buy a credit pack once and use it page by page. Nothing renews. You pick the quality per request.
Trial
- 3,000 credits for 30 days
- About 1,000 pages at the auto quality (3,000 fast pages)
- It does not turn into a subscription
| Quality | Credits per page | Good for |
|---|---|---|
| fast | 1 | Text and Markdown at speed, no models |
| balanced | 2 | Tables with the table model |
| auto | 3 | Picks the models a page needs. The everyday choice |
| max | 5 | Also describes figures and tables with a model |
max also describes figures and tables with a model: two per page are included and each one above that costs 1 more credit, up to six per page. With your own model key max costs the auto rate, 3.
A pack is one payment. When it is used up you buy the next one, or you do not.
No card, or need more volume? Email [email protected].
* The hosted API is coming soon and is not live yet. Its prices, credits, region and retention are the planned terms as of 16 November 2026 and are checked again at launch.
The small print,
up front.
If your documents cannot leave the building, use the SDK. It is the same engine, inside your own software.
- Where
- The API runs in the EU by default.
- Retention
- Synchronous requests run in memory. A job result, an upload through the upload URL, or an
output=urlresult is held for 15 minutes, then deleted. - Big jobs
- Long documents run as async jobs, with a webhook when they are done.
- Moving over
- There is an endpoint in the shape you already call for another parsing API, so you change a URL and compare.
- Your own key
- Pass your own model key for descriptions. With your own key max costs the auto rate, 3 credits per page.
- Agents
- The MCP tools match the API. The local MCP server runs on your own machine today.
Questions about Extract Hosted
Is the PDF extraction API available now?+
Can I try it first?+
Which quality should I use?+
Can my agent call it?+
What if my documents cannot leave the building?+
Be there when
it opens.
Extract Hosted is coming soon. The SDK runs the same engine today, inside your own software.