GLM OCR API

GLM OCR converts document images to Markdown or JSON for developers parsing invoices, forms and reports, from $0.027.

Image in, text out
8K token context
1.3B parameters, FP8
MIT licence

from openai import OpenAI

client = OpenAI(
    base_url="https://api.modelsheep.com",
    api_key=API_KEY_MODELSHEEP,
)

response = client.chat.completions.create(
    model="zai-org/GLM-OCR",
    messages=[
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Read the text in this image."
                },
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://modelsheep.com/invoice.png"
                    }
                }
            ]
        }
    ],
)

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

How to use the GLM OCR API

1
Create a ModelSheep account and copy an API key from the dashboard.
2
Point the OpenAI SDK at the base url https://api.modelsheep.com and set the model to zai-org/GLM-OCR.
3
Send one user message that holds a text part and an image part, pass the image as a URL or a base64 data URL, and read the Markdown or JSON back from the first choice.

What GLM OCR is

GLM OCR is a document recognition model from Z.ai. It reads a page image and returns the text, tables and formulas. Z.ai trained it for two tasks. It parses a document or extracts named fields. Other vision models in this catalogue answer open questions about an image. GLM OCR transcribes the page.

What it does well

Converts a scanned invoice or receipt to Markdown.
Rebuilds a table from a report page as HTML.
Extracts named fields from a form or certificate to JSON.

What it is not for

Open questions about a photo. The model card lists two tasks: document parsing and information extraction. The model does not describe a scene or discuss an image.
Long documents in one call. The context window holds 8K tokens. One dense page and a long JSON answer can fill it. Send one page per request.

Calling it

Call POST /chat/completions with the model string zai-org/GLM-OCR. Include a text part and an image part in the same user message. The image must be a URL or a base64 data URL. Pass a JSON schema in response_format when you need fields. A loose instruction returns loose text.

zai-org/GLM-OCR /chat/completions

SDKs and languages

GLM OCR answers on an HTTP endpoint, so any language that can post JSON reaches it. These are the ones the box above writes out.

OpenAI SDK: PythonTypeScript
Anthropic SDK: PythonTypeScript
HTTP: cURLPythonTypeScriptGoRustJavaC#PHP

Specifications

Author	Z.ai
Category	vision
Accepts	text, image
Returns	text
Total parameters	1.3B
Active parameters	dense, all of them
Score	not rated
Languages	8 namedChinese, English, French, Spanish, Russian, German, Japanese and Korean. Z.ai ends the list with "etc.", so eight is a floor. Arabic is not supported today.
Precision served	FP8
Context	8K tokens
Vector	n/a
License	MIT

Price

Input

$0.027

per million tokens

Output

$0.027

per million tokens

Against OpenRouter

10%

below the cheapest of 1 provider

The cheapest OpenRouter price for this model was $0.030 in and $0.030 out, with a median of $0.030 and $0.030 across the comparable providers. Read on 2026-09-17.

License

Z.ai publishes the weights under the MIT licence, which permits commercial use. The full Z.ai document pipeline adds the PP-DocLayoutV3 layout component under Apache 2.0. Read both licences before you self-host that pipeline.

Parameters: 304.2B, 13B
Context: 1.3M
Input per 1M: $0.027
Output per 1M: $0.117
License: MIT