GLM-OCR

Good fit for

Developers building document-recognition pipelines rather than looking for a general conversation assistant.

Why consider it

The official repository documents a 0.9B model and separate layout-analysis, recognition and formatting components.

Main limitations

The full PDF pipeline is more than the weights: SDK processing and layout detection are separate. Table recognition does not establish numerical Data Analysis. No File Uploads, general Code Execution or Web Search feature is assigned to model weights. Hosted MaaS API use sends documents to the provider and is outside this local-model listing.

Cost and access

No subscription or per-token hosted fee for local inference within permitted uses. Hardware, electricity and setup are separate. The model weights use MIT; repository code and the PP-DocLayout-V3 component use Apache 2.0. Comply with the terms of all components used.

Getting started

Install a current compatible Ollama version, review the exact license and run: ollama run glm-ocr. API refers to the local runtime endpoint, not a hosted API included with the weights. For image text recognition use: ollama run glm-ocr Text Recognition: ./image.png.