Overview

GLM-OCR reads complex documents, including tables, formulas, code-heavy pages and seals, and extracts structured information. It was trained with a multi-token prediction loss and reinforcement learning, and Zhipu reports a score of 94.62 on OmniDocBench V1.5, first overall at release.

Highlights

  • 0.9B parameters
  • Formula, table and information extraction
  • Supports vLLM, SGLang and Ollama
  • SDK and inference toolchain included
  • MIT license

Typical uses

Document digitization, invoice and form processing, and preparing PDFs for RAG.

Things to consider

  • Benchmark results are the developer’s own.
  • Very unusual layouts may still need a full vision-language model.
  • Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.

Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.