Overview
GLM-Image is an open-weight image-generation model designed for instruction-following and knowledge- or text-dense visuals. It combines a 9B-parameter autoregressive generator with a 7B-parameter diffusion decoder. Zhipu says the model was trained throughout on domestically produced chips. (docs.bigmodel.cn)
Release date: January 14, 2026, according to the official release record. (docs.bigmodel.cn)
Capabilities and uses
- Generates images from text, with particular emphasis on rendering text, including Chinese characters, and following complex instructions.
- The open model also supports image-to-image tasks, including editing, style transfer, identity preservation and multi-subject consistency. These capabilities should not be assumed to be available through every hosted API.
- Intended uses include commercial posters, educational illustrations, diagrams, multi-panel images and social-media graphics. (docs.bigmodel.cn)
Limits and deployment
- The domestic API documentation specifies a maximum text input of 1,000 characters. Custom output width and height must each be 512–2,048 pixels and divisible by 32. The API returns an image URL for download.
- The repository recommends enclosing text intended to appear in the image in quotation marks. Generated text and factual content should still be checked.
- For local inference, the repository notes substantial compute requirements and recommends a GPU with more than 80 GB of memory or a multi-GPU setup. Its code repository is published under Apache 2.0. (docs.bigmodel.cn)
Official API pricing
Prices checked on November 17, 2026. The domestic and international prices apply to their respective platforms and are not currency conversions of one another.
- Mainland China — BigModel: CNY 0.1 per request for multi-resolution image generation. The official table lists Batch API as unsupported. It does not give separate input-token and output-token rates for this model. Domestic pricing. (docs.bigmodel.cn)
- International — Z.ai: USD 0.015 per generated image. The official table does not give separate input-token and output-token rates for this model. International pricing. (docs.z.ai)