One page is one page. A precise page is twenty.
No per-word pricing and no surprise overage bill. Fast and auto cost a page a page; the precise engine spends a vision model's full attention on every page, and is priced like it. When you hit the quota the API says so and stops.
Quotas reset 00:00 UTC on the 1st · unused pages do not roll over · failed reads are never billed · a precise page draws 20
What counts as a page?
One recognised page. An image is one page; a PDF is one page per page, so a six-page statement costs six. A dense A4 scan and a cropped screenshot cost the same. Requests that fail recognition are logged for you to inspect but never billed — and a PDF that exceeds your plan's per-document limit is rejected before any of it is charged.
Does the precise engine cost more?
Yes — one precise page draws 20 pages of your allowance, and it needs Pro
or above. It is not a margin grab: a fast page takes about a second and a
half of a CPU, a precise page takes about fifty seconds of a vision model.
Charging both the same would mean either capping the engine far lower or
pricing the plan for the worst case. The response tells you the rate in
page_multiplier, and a request that would exceed
the quota is refused before any work is done.
What happens when I hit the quota?
Calls return 402 quota_exceeded until you upgrade or the month
turns over. Nothing silently degrades and nothing auto-bills.
Can I self-host it?
Both engines run locally — Tesseract and a local vision model, no third-party API in the path. Your documents are not sent anywhere else. Get in touch about running it inside your own network.