On February 3, 2026, Zhipu AI made a significant move by officially open-sourcing its professional-grade OCR (Optical Character Recognition) model, GLM-OCR. This model is remarkable for its lightweight design, achieved with just 0.9 billion parameters. It has excelled in the authoritative document parsing benchmark, OmniDocBench V1.5, scoring an impressive 94.6 and thus securing the top position for current performance. GLM-OCR offers flexible deployment options via vLLM, SGLang, and Ollama. It shines in tasks like formula recognition, table recognition, and information extraction, with some of its capabilities coming close to those of the versatile large model Gemini-3-Pro.
