Unstructured
Unstructured pricing charges $0.015 for every page it processes, the same whether that page is a clean text PDF or a scanned table running through generative OCR.
Updated on:
Unstructured pricing: one rate, and a page that isn't always a page
Unstructured pricing charges $0.015 for every page it processes, the same whether that page is a clean text PDF or a scanned table running through generative OCR. Firecrawl's one credit per page holds until you attach a modifier, and then the bill multiplies. Unstructured has no modifiers. The variable sits one layer down, in what counts as a page.
Key takeaways
Unstructured bills $0.015 a page processed, flat, with 10,000 free pages granted once per account rather than every month.
The rate doesn't move with strategy or enrichment. Fast, High-Res, VLM, generative OCR, table description and named entity recognition all cost the same per page.
Outside PDF, PPTX, TIFF and page-tagged DOCX, a page means 100 KB of file size, so a 10 MB JSON export bills as roughly 102 pages.
The pricing page's own structured data still advertises the retired scheme, $0.03 a page and 15,000 free pages a month, double the rate the page shows a human.
Unstructured pricing in 2026
Plan | Price | Pages | Deployment |
|---|---|---|---|
Free | $0, no card | 10,000, once per account | Shared SaaS |
Pay-As-You-Go | $0.015/page | Unlimited after the free 10,000 | Shared SaaS |
Business | Custom | Custom, volume discounts by quote | Dedicated, VPC or bare metal |
What Unstructured actually meters
Unstructured meters pages, and publishes exactly how it counts them.
For PDF, PPTX and TIFF, a page is a page, a slide or an image. For DOCX carrying page metadata, it reads the count from there. For every other file type, it divides the file's size by 100 KB, and for non-file connector sources, 100 KB of incoming data is one page.
That last rule is where per-page billing stops matching intuition. A 5 MB spreadsheet has no pages at all, and Unstructured bills it as roughly 51. A 10 MB JSON export costs about $1.54. Estimating from document counts models the PDF half of a corpus correctly and the rest not at all.
What doesn't vary is the rate. Fast, High-Res, VLM and Auto strategies cost the same. Generative OCR, image description, table description and named entity recognition cost the same. Chunking and embedding cost the same. There's no premium tier of the parser, so the only lever on your bill is how many pages you send.
Two things stay undocumented. Volume discounts exist with no published threshold or rate. And the docs note that the Billing tab "shows your account's current billing rate per page", conceding the rate is per-account and visible only once you sign in.
How the free tier works
The 10,000 free pages are a one-time account credit, not a monthly allowance.
The docs say it plainly: Unstructured applies the credit once per account, and once it's gone, requests begin failing until you upgrade. No card is needed to claim it, and nothing resets.
At $0.015 that credit is worth $150, which buys a real evaluation rather than a demo. The older scheme was more generous: until June 2026 the page offered 15,000 free pages at $0.03, worth $450, with no expiration date.
There's no balance to top up, no rollover question and no deduction order, because there are no credits here in the billing sense. You use the grant, then pay per page in arrears.
What happens when you hit the limit
Requests fail with HTTP 402 Payment Required.
Unstructured's billing doc states that after the first 10,000 pages, requests begin failing unless you've upgraded to Pay-As-You-Go. The support docs list the same 402 beside Insufficient API quota errors, and recommend slowing the request rate, batching smaller or upgrading.
Once a card is attached there's no second cliff. Pay-As-You-Go bills each calendar month with no maximum and no stated spend cap, so a runaway pipeline over a 10 million page corpus bills $150,000 without stopping. Nothing in the docs describes a budget limit or an alert.
VPC deployments carry a cost Unstructured doesn't bill. Its support page notes that some of those plans require the customer to pay for related services and models directly, so the vector database and the model calls arrive on separate invoices.
How Unstructured's pricing has changed
Date | Milestone | Source |
|---|---|---|
Jun to Oct 2026 | Rate halves to $0.015, free grant drops from 15,000 to 10,000 pages | Vendor |
Jun to Oct 2026 | Free pages become a one-off account credit, not an open grant | Vendor |
18 Nov 2025 | Page shows $0.03 a page and 15,000 free pages, "no expiration date" | Vendor |
Jan 2025 to Oct 2026 | The 100 KB page definition stays unchanged, word for word | Vendor |
Customer
Sentiment Highlights
"Unstructured.io, one of your competitors, goes as far as having an AWS Marketplace setup so customers can use their own infrastruture but still pay them."
Developer comparing document-parsing vendors, Hacker News, December 2024
"Setting up the open-source, locally hosted version of Unstructured was quite cumbersome due to the numerous additional packages and installation steps required."
Developer extracting information from PDFs, Hacker News, November 2024
Explore other providers
Firecrawl
Developer Tool
Firecrawl pricing runs on a single credit, and one credit buys one page on a basic scrape, crawl or map.
Fireworks AI
Infrastructure Platform
Fireworks AI pricing runs two meters: per token on serverless, and per GPU-second on on-demand deployments.
Modal
Infrastructure Platform
Modal pricing charges by the second for three resources at once: CPU cores, memory and GPU each carry their own rate, and a single function's bill is the sum.
How much does Unstructured cost per page?
Does Unstructured charge extra for OCR or table extraction?
How does Unstructured count a page?
Can you self-host Unstructured?
