Overview
GLM-OCR is a lightweight, 0.9-billion-parameter multimodal model specialized in optical character recognition and document parsing. Its encoder–decoder design combines a CogViT visual encoder with a GLM-0.5B language decoder. The full parsing pipeline uses layout analysis followed by parallel region recognition. (docs.bigmodel.cn)
Release date: February 3, 2026, according to Zhipu AI’s official release record. (docs.bigmodel.cn)
Capabilities and uses
- Recognizes printed text, handwriting, mathematical formulas, code, seals and complex tables; it can extract information from documents such as receipts, forms and certificates.
- Supports document-processing workflows, data entry and retrieval-augmented generation. Outputs include text, Markdown documents, image links and structured JSON; table content can be represented as HTML.
- Supports multiple languages, including Chinese, English, French, Spanish, Russian, German, Japanese and Korean.
- The publisher reports a score of 94.62 on OmniDocBench V1.5 at launch. This is a benchmark result, not a guarantee of accuracy on every document. (docs.bigmodel.cn)
Availability and limitations
The hosted documentation specifies PDF and JPG/PNG inputs, with a maximum of 10 MB per image, 50 MB per PDF and 100 pages. Actual performance depends on factors including file quality, network conditions and concurrency; important extracted content should be checked. These hosted-service limits should not be assumed to apply unchanged to self-hosted deployments. (docs.bigmodel.cn)
The model is available for self-hosting, including with vLLM, SGLang and Ollama. The model is released under the MIT License; the repository code and the PP-DocLayoutV3 component used in the complete pipeline are under Apache 2.0. Self-hosting entails separate infrastructure costs. (github.com)
Official API pricing
Prices checked November 16, 2026. These are separate regional platform prices, not currency conversions or necessarily identical packages.
- Mainland China — BigModel: CNY 0.2 per 1 million input tokens and CNY 0.2 per 1 million output tokens. The rate is stated in the official Chinese GLM-OCR documentation; the linked BigModel pricing page could not be read without JavaScript. (docs.bigmodel.cn)
- International — Z.AI: USD 0.03 per 1 million input tokens and USD 0.03 per 1 million output tokens, according to the official Z.AI pricing page. Its GLM-OCR row does not list a cached-input rate. (docs.z.ai)