> ## Documentation Index
> Fetch the complete documentation index at: https://help.abacusdocs.com/llms.txt
> Use this file to discover all available pages before exploring further.
# What gets billed
> Pages vs model usage: what each product surface spends, and what is free.
Abacus Docs charges **credits** from your organization's balance. One credit is **\$0.01** USD. The header chip and **Settings → Billing** show what you have left. **Settings → Usage** shows who spent it.
There are two meters, not one:
| Meter | Used for | How it is priced |
| --------------- | --------------------------------------------------------------- | --------------------------------------------------------------------- |
| **Pages** | Every extraction that produces a transcription or schema fields | A fixed number of credits **per page**, once the run completes |
| **Model usage** | Schema-builder AI, Extract Chat, and Deep Reader | Tokens for that turn. Chat adds a **1.5×** markup on the model's cost |
You are not charged for opening a document, searching the library, editing a cell, exporting, or sharing.
## Page rates
A *page* is one page of the file after we accept it (a multi-page PDF is N pages; a folder of images becomes one PDF and is billed on that page count).
| Your organization | Single document | Batch |
| ---------------------------------------------------------- | -------------------- | -------------------- |
| Individual | **4** credits / page | **7** credits / page |
| Enterprise, prepaid balance remaining | **3** / page | **6** / page |
| Enterprise, invoiced in arrears or prepaid balance at zero | **4** / page | **7** / page |
**Single document** is anything that does not go through **Batch Processing**: My Documents, a file attached in Chat, a schema sample, a Full Test, an API upload.
**Batch** is a job on **Batch Processing**, a [Batch Full Test](/schemas/publish-and-versions) in the builder, and [Black Hole](/black-hole) files after they are filed into a schema's default batch.
The upload form estimates this *before* processing starts and can block the upload if the estimate would exceed your balance. Enterprise organizations skip that gate. Chat questions are not estimated in advance.
## What spends pages
Each **completed** extraction is a charge. The same file can be charged more than once if you run it again.
Dropping a PDF or image and clicking **Begin Document Extraction** bills the page rate for a single document when processing finishes. Applying a schema later and clicking **Reprocess** is a **new** charge.
Each file that completes in the job is billed at the **batch** page rate. Completed files are not billed again if you retry only the failures. Adding files or reprocessing selected documents is a new charge on those files.
**Reprocess** on a finished document is a new page charge (single-document or batch rate, matching where it lives). [Moving a document](/guides/exporting-results#move-a-document) to another batch always re-extracts and bills the mover, even when the schema is the same.
The sample you upload in the [builder](/schemas/builder), and each PDF you add to the schema's **test set**, is billed like a My Documents upload (single-document page rate). The file is extracted so later tests have something to run against.
**Full Test** re-runs the *draft* schema against the sample (and test-set files). That extraction is billed again at the single-document page rate. It does not overwrite your library documents.
A Batch Full Test runs the draft against the whole test set as a hidden batch. Each completed file is billed at the **batch** page rate. Your real batches are not touched.
Classifying a dropped file is not a separate credit line. Once the file is filed into a schema's default batch (automatically or by a mod **Route**), extraction is billed at the **batch** page rate, attributed to the person whose agent dropped the file.
A document posted with an organization API key is billed at the single-document page rate against that organization, on the key.
A run that **fails** is not charged per page. **Retry** on a failed document resumes; it is not priced as a full reprocess. **Reprocess** on a completed document is.
## What spends model usage
These do **not** use the page table. Cost depends on how long the model ran on that turn.
### Schema builder agent
The AI column on a schema — ordinary chat and **Auto-improve** — is billed from the model's token usage for that run. There is **no** 1.5× chat markup on the builder.
Adding or editing fields by hand, changing mode, and clicking **Publish** do not spend credits.
### Extract Chat and Deep Reader
Questions in sidebar **Chat** and on a document's **Chat** tab are billed from that turn's model usage, at **1.5×** the model's cost. A \$1.00 model bill becomes **150 credits**.
Attaching or uploading a file in Chat still **extracts** it first. That extraction is a normal page charge; the questions you ask afterwards are the chat charge.
Asking a statistical question over an attached batch (counts, totals, filters) runs a read-only query on the grid. The query itself does not spend credits. The model's reply around it does.
## What does not spend credits
* Opening **My Documents**, the viewer, or a batch grid
* Library **search**
* Editing a cell (overrides)
* Exporting CSV, Excel, JSON, or an archive
* Creating or revoking a [share link](/sharing/public-links)
* Changing a member's [role](/sharing/members)
* Saving a schema draft or publishing a version (the tests that precede publish are billed; the publish click is not)
## Who is charged
The **organization** pays. **Settings → Usage** names the person who triggered the work: the uploader, whoever clicked Reprocess, or whoever sent the chat message. For Black Hole, spend is attributed to the person whose agent dropped the file, even when a moderator later routes it.
A teammate's shared batch still bills your organization when *you* reprocess or chat against it. The usage row is yours.
## Worked example
An individual account uploads a 10-page invoice on My Documents: **40 credits**. They design a schema, talk to the builder agent (model usage, no markup), then run Full Test: **another 40 credits**. They put 200 similar invoices through Batch Processing: **1,400 credits** (200 × 10 × 7). They open the batch in Chat and ask who is late: that turn is model usage at 1.5×, not another 1,400.
See [Credits and billing](/account/credits) for packs, the free grant, and what happens at zero.