GLM-OCR is a 0.9B multimodal OCR model that achieves 94.62 on OmniDocBench V1.5, ranking first overall despite its compact size. Built by Zhipu AI, it beats Qwen3-VL-235B (260× more parameters) and Gemini-3 Pro on document parsing benchmarks.
The model combines a 0.4B CogViT visual encoder with a 0.5B GLM language decoder. Its key innovation is Multi-Token Prediction (MTP), which predicts 10 tokens per step instead of one. For OCR—a deterministic task—you’re copying characters, not sampling creative text. This delivers about 50% higher throughput vs standard decoding.
The full pipeline uses PP-DocLayout-V3 for layout detection, then runs parallel OCR across different document regions. It achieves 1.86 pages/second for PDFs and 0.67 images/second, running in just 1.5GB VRAM.
Deployment options include Zhipu MaaS API (no GPU needed), self-hosted vLLM/SGLang, or Ollama. The SDK supports CLI (glmocr parse image.png), Python API, or Flask service.