PaddlePaddle

PP-OCRv6

About

PP-OCRv6 is a 20M open-weight model from PaddlePaddle. PaddleOCR PP-OCRv6 medium text detection and recognition; scene OCR JSONL on image chat; text returns plain text; document_url fan-out via text. It accepts images and documents. Released June 11, 2026.

Parameters
20M
Published parameter count.
License
Apache-2.0
License on the weights.

Pricing

Token rates are US dollars per 1M tokens. Image rates are per 1K images. Audio rates are per 1M audio seconds.

Rate	Price
Input / 1M	$0.02
Cached input / 1M	$0.02
Output / 1M	$0.40

Compare

PP-OCRv6 is the smallest of these OCR models, at 20M, and its published input rate is the lowest, $0.02 / 1M input.

Model	What it does	Price	Parameters	Context
PP-OCRv6	PaddleOCR PP-OCRv6 medium text detection and recognition; scene OCR JSONL on image chat; text returns plain text; document_url fan-out via text.	$0.02 / 1M input	20M	—
PaddleOCR-VL 1.6	PaddleOCR-VL-1.6 for OCR, tables, formulas, and charts.	$0.15 / 1M input	0.9B	16K
DeepSeek-OCR-2	DeepSeek OCR 2 and markdown extraction.	$0.25 / 1M input	3.4B	32K
dots.mocr	Multilingual document layout parsing and markdown OCR.	$0.20 / 1M input	3B	32K

Benchmarks

Published scores for PP-OCRv6, from the PP-OCRv6 paper. Medium tier.

Evals

Reading	Example
Qualitative	Clear structure, grounded in the input
Quantitative	Detection Hmean: 86.2.
Cost and performance	Lower listed rate, mid-pack latency

Performance

Latency is the end-to-end round trip. Throughput is completion tokens divided by that time. It is a request rate, not decode speed.

Latency
139 ms
Throughput
536 tok/s
Requests
7,678

Examples

Scene-text OCR
PP-OCRv6 reads a New York Times newsstand in 120 ms.
{"lines":[{"text":"The New York Times","score":0.99,"box_xyxy":[..]}, ...]}

Supported inputs

textNo
A request with no media part is a 400.
image_urlYes
One per request. A second image_url part is a 400.
video_urlNo
Not accepted.
document_urlYes
One PDF, up to 128 pages, rasterized per page at document_dpi (default 96). See Document Inputs.

Default method ocr.
Alias: pp-ocrv6.

Methods

Methods this model serves. Payload shapes are in the docs.

Method	Returns
ocr	Region items with `bbox_xywh`, optional `poly_xy`, `text`, `score`
detect	Region items with `bbox_xywh` and optional `poly_xy`, geometry only
text	Recognized text stitched in reading order

The payload kind is the same for an image and for each page of a PDF. Text mode emits the payload alone for one image, and one <document> / <page> block for a PDF. In JSON mode an image content is the tagged container above, and a document page content is always doc.page.blocks. See Methods & Response Format.

Coordinates are normalized 0-1 against the image or rasterized page, at precision decimals (default 4). text is plain scene text stitched in reading order, not generative Markdown.

Method parameters

Parameter	Default	Description
`lang`	`en`	Recognition language (`en`, `ch`, `japan`, ...), on every method.
`score_threshold`	`0.5`	`ocr` and `text`: drop text regions below this recognition confidence (0.0-1.0). `detect` returns every region.

Estimate cost

Estimate. A page is 2,500 input tokens and 1,000 output tokens.

Pages

$0.45

Quick start

Call this model on the OpenAI-compatible gateway. The model id is already filled in.

from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.vlm.run/v1/openai",
    api_key="<VLMRUN_API_KEY>",
)

response = client.chat.completions.create(
    model="paddleocr/pp-ocrv6",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://storage.googleapis.com/vlm-data-public-prod/hub/examples/document.receipt/playground/2.jpg"
                    },
                },
            ],
        }
    ],
    extra_body={"method": "ocr"},
)

print(response.choices[0].message.content)

Response

A single image returns the region container alone, with no wrapper:

{"object": "doc.ocr.lines", "items": [{"bbox_xywh": [0.0332, 0.0138, 0.1719, 0.0331], "poly_xy": [[0.0332, 0.0138], [0.2051, 0.0138], [0.2051, 0.047], [0.0332, 0.047]], "text": "Invoice", "score": 0.998}, {"bbox_xywh": [0.0332, 0.0829, 0.2598, 0.0331], "text": "Total: $99.00", "score": 0.987}]}

For PDFs, each page's content is doc.page.blocks, and every region becomes one block with the same geometry under a zero-based block_id. See the Document OCR guide.