> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reducto.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Credit Usage
> Understanding how credits are calculated across Reducto endpoints
Credits are the billing unit for Reducto API usage. This page explains how credits are calculated for each endpoint and document type.
## Parse Endpoint
### Documents and Images
Supported formats: PDF, DOCX, DOC, PPTX, PPT, PNG, JPEG, GIF, TIFF, HEIC, and more.
| Processing Type | Credits/Page | When Applied |
| ------------------------ | ------------ | ---------------------------------------------------------------------------------- |
| **Standard** | 1 | Text, layout, simple tables, OCR |
| **Complex** | 2 | VLM-enhanced pages (complex tables, key-value regions, figures with summarization) |
| **Agentic - Standard** | 2 | Agentic mode enabled on simple pages |
| **Agentic - Complex** | 4 | Agentic mode enabled on complex pages |
| **Advanced chart agent** | +4 | Per chart when `advanced_chart_agent: true` |
Reducto automatically classifies page complexity. You don't choose "standard" vs "complex". Complexity is determined by the content.
**What makes a page complex?**
* Tables with merged cells or nested headers
* Key-value form regions
* Figures when `summarize_figures: true` (default)
### Spreadsheets
Supported formats: XLSX, XLS, CSV, XLSM
| Clustering Mode | Credits | Description |
| ---------------------- | ----------------- | --------------------------- |
| **Accurate** (default) | 1 per 1,000 cells | Intelligent table detection |
| **Fast** | 1 per 5,000 cells | Basic clustering |
| **Disabled** | 1 per 5,000 cells | Single table output |
### Text Files
Supported formats: HTML, TXT, RTF
| Format | Credits/Page |
| ------------ | ------------ |
| Text formats | 0.5 |
***
## Extract Endpoint
| Mode | Credits | Notes |
| ----------------------- | ------------------------------------------------------------------------------- | -------------------------------------- |
| **Standard** | 2 per page | Schema-based extraction |
| **Deep Extract** (Beta) | 4 per page + 0.1 per field extracted,
minimum of 30 credits per document\* | Agentic loop for near-perfect accuracy |
\*Beta pricing, subject to change.
If you pass a URL or file directly to Extract (instead of a `jobid://`), Parse credits are also charged. To avoid double-charging, parse first, then extract using the job ID.
```python theme={null}
# ❌ Charges for both Parse AND Extract
result = client.extract.run(
input=upload.file_id,
instructions={"schema": schema}
)
# ✅ Only charges for Extract (reuses parsed content)
parse_result = client.parse.run(input=upload.file_id)
result = client.extract.run(
input=f"jobid://{parse_result.job_id}",
instructions={"schema": schema}
)
```
***
## Split Endpoint
| Mode | Credits/Page |
| ------------ | ------------ |
| **Standard** | 2 |
Same as Extract: passing a URL charges Parse + Split. Use `jobid://` to avoid double-charging.
***
## Classify Endpoint
| Mode | Credits/Page | Notes |
| ------------ | ------------ | ------------------------ |
| **Standard** | 0.5 | Per page of context used |
Classify reads the first 5 pages of a document by default, costing 2.5 credits per classification. You can configure context up to 10 pages (5.0 credits).
***
## Edit Endpoint
| Mode | Credits/Page | Notes |
| -------- | ------------ | ----------------- |
| **Beta** | 4 | Subject to change |
***
## Pipeline Endpoint
Pipelines combine multiple operations. Credits are the sum of all operations in the pipeline:
```
Pipeline credits = Parse + Extract (if configured) + Split (if configured)
```
***
## Credit Optimization Tips
### 1. Reuse Parse Results
Parse once, then run multiple Extract or Split calls using the job ID:
```python theme={null}
# Parse once
parse = client.parse.run(input=upload.file_id)
# Extract multiple schemas without re-parsing
invoice_data = client.extract.run(
input=f"jobid://{parse.job_id}",
instructions={"schema": invoice_schema}
)
vendor_data = client.extract.run(
input=f"jobid://{parse.job_id}",
instructions={"schema": vendor_schema}
)
```
### 2. Disable Agentic Mode When Not Needed
Agentic mode doubles credit usage. Only enable it for:
* Handwritten content
* Low-quality scans
* Complex tables that parse incorrectly
```python theme={null}
# Standard parsing (1-2 credits/page)
result = client.parse.run(input=upload.file_id)
# Only enable agentic when needed (2-4 credits/page)
result = client.parse.run(
input=upload.file_id,
enhance={"agentic": [{"scope": "text"}]}
)
```
### 3. Use Page Ranges
Process only the pages you need:
```python theme={null}
# Only process pages 1-10
result = client.parse.run(
input=upload.file_id,
settings={"page_range": {"start": 1, "end": 10}}
)
```
### 4. Choose Appropriate Spreadsheet Clustering
For large spreadsheets where you don't need intelligent table detection:
```python theme={null}
result = client.parse.run(
input=upload.file_id, # spreadsheet file
spreadsheet={"clustering": "fast"} # 5x cheaper than "accurate"
)
```
***
## Monitoring Usage
Track your credit usage in [Reducto Studio](https://studio.reducto.ai/):
* **Usage dashboard**: View credits consumed over time
* **Job logs**: See credits charged per job
* **Usage alerts**: Set threshold notifications
Each API response also includes credit information:
```json theme={null}
{
"job_id": "abc123",
"usage": {
"num_pages": 10,
"credits": 15.0
}
}
```
***
## Related
Request limits and optimization.
Reuse parse results to save credits.