Token rates
Each model is billed for the tokens it reads and the tokens it writes. Prices are per million tokens.
OCR reads the image or document and transcribes the text on the page. POST /ocr returns that transcription as Markdown and bills only the OCR model.
Endpoints that return structured data run a second step. The language model reads the transcription and turns it into JSON: named fields on POST /ocr.json and POST /extract-form-values.json, and document classification on POST /paperless. Those requests bill both models. On POST /paperless, the OCR model runs only when advanced OCR is on.
| Endpoint | Model | Input | Output |
|---|---|---|---|
POST /ocr |
OCR | $0.40 | $0.70 |
POST /ocr.json |
OCR | $0.40 | $0.70 |
POST /ocr.json |
Language model | $0.20 | $1.60 |
POST /extract-form-values.json |
OCR | $0.40 | $0.70 |
POST /extract-form-values.json |
Language model | $0.20 | $1.60 |
POST /paperless |
OCR, only when advanced OCR is on | $0.40 | $0.70 |
POST /paperless |
Language model | $0.20 | $1.60 |
Estimate from a page count
One page is 2,900 OCR input tokens and 800 OCR output tokens. A language model adds 5,500 input tokens and 1,400 output tokens.
POST /ocr
$0.001720
OCR: 2,900 input, 800 output
POST /ocr.json
$0.005060
OCR: 2,900 input, 800 output. Language model: 5,500 input, 1,400 output.
POST /extract-form-values.json
$0.005060
OCR: 2,900 input, 800 output. Language model: 5,500 input, 1,400 output.
POST /paperless
Language model: 5,500 input, 1,400 output
$0.003340
With advanced OCR, add 2,900 OCR input and 800 OCR output
$0.005060