GLM-OCR is an open-source optical character recognition model from Z.ai, the team behind the GLM family of language models. Built as a vision-language model, it reads text in images and documents and returns accurate, structured transcriptions — summed up by the project as “accurate, fast, comprehensive”. The repository has about 7,500 stars on GitHub, is released under the Apache 2.0 license, and was actively updated in 2026.
Traditional OCR engines treat a page as pixels to be decoded character by character, and they struggle with messy layouts, mixed languages and low-quality scans. GLM-OCR approaches the problem with a modern multimodal model: the same class of architecture that powers general vision-language systems, specialized for document understanding. That makes it suitable for the documents businesses actually process — invoices, forms, receipts, reports, screenshots and photographed pages — where reading order and structure matter, not just characters. The repository ships the model code and weights for self-hosted inference, so document pipelines can run OCR inside their own infrastructure instead of sending sensitive pages to a third-party API.
Free. GLM-OCR is open source under the Apache 2.0 license; you run the model on your own hardware at no licensing cost.
Developers building document-processing pipelines — digitization, invoice automation, archival search and RAG ingestion — who want modern multimodal OCR they can host and control themselves.