DeepSeek

DeepSeek-OCR-2

About

DeepSeek-OCR-2 is a 3.4B open-weight model from DeepSeek. DeepSeek OCR 2 and markdown extraction. It accepts text, images, and documents. Context window is 32K. Released January 27, 2026.

Parameters
3.4B
Published parameter count.
Context window
32K
Tokens of context on a request.
License
MIT
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	$0.25
Cached input / 1M	$0.25
Output / 1M	$0.80

Compare

DeepSeek-OCR-2 is one of the catalog's OCR models. The table is each model's published price, size, and context.

Model	What it does	Price	Parameters	Context
DeepSeek-OCR-2	DeepSeek OCR 2 and markdown extraction.	$0.25 / 1M input	3.4B	32K
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	—
dots.mocr	Multilingual document layout parsing and markdown OCR.	$0.20 / 1M input	3B	32K
GLM-OCR	Compact multilingual document OCR and markdown extraction.	$0.10 / 1M input	0.9B	8K

Benchmarks

Published scores for DeepSeek-OCR-2, from the OmniDocBench leaderboard.

Evals

Reading	Example
Qualitative	Clear structure, grounded in the input
Quantitative	OmniDocBench v1.6: 90.25.
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
719 ms
Throughput
—
Requests
151

Supported inputs

textIgnored
A text part beside the media is accepted and ignored. 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 markdown. Three methods return the page as a string; grounding_ocr locates a phrase and returns region boxes instead.
Context window 32,768 tokens, with a maximum output of 8,192 tokens.
Weights: deepseek-ai/DeepSeek-OCR-2. Aliases are listed below.

Methods

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

Method	Returns
markdown	Reading-order Markdown string (default)
ocr	Markdown via the grounded conversion prompt; grounding tags are stripped from the output
free_ocr	Free-form plain read of the page
grounding_ocr	Region items with `bbox_xywh` and the matched `text`

Text mode emits the string alone for one image, and one <document> / <page> block per PDF page. In JSON mode an image content is the string, or the region container for grounding_ocr, and a document page content is always doc.page.blocks. See Methods & Response Format.

The model grounds on a fixed 0-1000 grid; the Gateway normalizes every box to 0-1 against the image or rasterized page, at precision decimals (default 4). Grounding regions carry bbox_xywh and text only: no polygon and no score.

Method parameters

Parameter	Default	Description
`prompt`	none	`grounding_ocr` only: the full grounding prompt, with the phrase to locate wrapped in the model's ref markers. See Grounding.

Grounding

grounding_ocr answers "where does this appear on the page". Pass the full grounding prompt via method_params.prompt, with the phrase to locate wrapped in the model's ref markers:

response = client.chat.completions.create(
    model="deepseek-ai/deepseek-ocr-2",
    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": "grounding_ocr",
        "method_params": {"prompt": "<image>\nLocate <|ref|>Total Amount<|/ref|> in the image."},
    },
)

The reply is the region container, alone in text mode and under content in JSON mode:

{"object": "doc.ocr.lines", "items": [{"bbox_xywh": [0.102, 0.08, 0.893, 0.06], "text": "Annual Report 2024"}, {"bbox_xywh": [0.102, 0.16, 0.893, 0.08], "text": "Overview"}]}

On a PDF each page's content is doc.page.blocks, with one block per located region carrying the same bbox_xywh and text.

Aliases

deepseek-ai/deepseek-ocr-2 (preferred), deepseek-ocr-2, and deepseek-ai/DeepSeek-OCR-2 all resolve to this model.

Estimate cost

Estimate. A page is 1,200 input tokens and 1,000 output tokens.

Pages

$1.10

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="deepseek-ai/deepseek-ocr-2",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "document_url",
                    "document_url": {
                        "url": "https://storage.googleapis.com/vlm-data-public-prod/hub/examples/finance.sec-filings/tsla-8k.pdf"
                    },
                },
            ],
        }
    ],
    extra_body={"method": "markdown"},
)

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

Response

One <document> block per PDF, one <page> block per page, each carrying format="markdown":

tsla-8k.pdf2 pages

Page 0

Annual Report 2024

Overview

The company reported revenue of $1.2M.

Page 1

Metric	2023	2024
Revenue	0.9M	1.2M

A single image returns the Markdown string alone, with no wrapper.