GLM-OCR
Zhipu AI
GLM-OCR is Z.ai's compact OCR and document-parsing model, launched 3 February 2026. It pairs the in-house CogViT visual encoder with a GLM-0.5B language decoder through a lightweight cross-modal connector that downsamples visual tokens, totalling 0.9B parameters, and is trained with a multi-token-prediction loss and full-task reinforcement learning. Weights are MIT-licensed on Hugging Face and it also runs as a hosted API under the model id glm-ocr. Prompting is restricted to text, formula and table recognition plus information extraction against a caller-supplied JSON schema. The official SDK adds PP-DocLayoutV3 layout analysis for end-to-end parsing.
Strengths
- Scores 94.62 on OmniDocBench V1.5, which the vendor reports as #1 overall
- Only 0.9B parameters, with vLLM, SGLang and Ollama deployment paths (ollama run glm-ocr)
- MIT-licensed weights alongside a hosted API priced identically for input and output at $0.03 per million tokens
- Vendor-measured throughput of 1.86 pages/second on PDFs and 0.67 images/second, under single-replica single-concurrency conditions
- Information-extraction mode returns data conforming to a caller-supplied JSON schema, suited to cards, certificates and forms
- Card declares Chinese, English, French, Spanish, Russian, German, Japanese and Korean
- Official SDK bundles PP-DocLayoutV3 layout analysis into a complete document-parsing pipeline with structured output
Caveats
- Prompting is explicitly limited to text/formula/table recognition and JSON-schema information extraction, so this is not a general vision-language model
- The SDK covers document parsing only; information-extraction tasks require calling the model directly, without the layout pipeline
- The end-to-end pipeline depends on PP-DocLayoutV3 under Apache 2.0, so two licences apply even though the model itself is MIT
- Apart from the single 94.62 headline, the card's performance sections are chart images only, so no per-task scores are recorded here
- The hosted API caps input at 10MB per image, 50MB per PDF and 100 pages per request
- Vendor
- Zhipu AI
- License
- MIT
- Release Date
- 2026-02-03
- Modalities
- textimage
| Benchmark | Score | Max |
|---|---|---|
| OmniDocBench V1.5 | 94.6 | 100.0 |
Capabilities
Resources
Pricing
Hosted API model id glm-ocr is priced at $0.03 per 1M input tokens and $0.03 per 1M output tokens on Z.ai's pricing page, with no cached-input tier offered. Weights are separately available under the MIT licence for self-hosting.