GLM OCR

GLM-OCR is a multimodal image-to-text OCR model from zai-org built on a GLM-V encoder-decoder architecture. It uses a CogViT visual encoder, a cross-modal connector with token downsampling, and a GLM-0.5B language decoder, with Multi-Token Prediction loss and full-task reinforcement learning. The model card reports support for languages including Chinese, English, French, Spanish, Russian, German, Japanese, and Korean, and describes a two-stage layout analysis and parallel recognition pipeline based on PP-DocLayout-V3. It is released under the MIT license with Safetensors tags and lists Transformers as the library name. The card states it has 0.9B parameters and supports deployment via vLLM, SGLang, and Ollama, and reports a score of 94.62 on OmniDocBench V1.5 plus throughput results under specified conditions.