Back to Gap Map Model components / Image, video, 3D & music generation

GLM-Image

Z.ai
open weights / Overall score: 2.0

GLM-Image is Z.ai's open image generation model, released in January 2026. A 9B autoregressive generator built from GLM-4-9B lays out the image as visual tokens and a 7B diffusion decoder renders them, which makes it strong at legible text inside images and at knowledge-heavy prompts such as posters and diagrams. The same model edits images, transfers styles and keeps subjects consistent across images.

The weights are MIT-licensed and the repository code Apache 2.0; the vision tokenizer and encoder weights it carries come from X-Omni under Apache 2.0. Z.ai also serves GLM-Image through its API. CogView, Z.ai's earlier image line, is not part of this entry.

Openness

3 high confidence
3.0
weights
open (the GLM-Image checkpoint downloads from the Hub without a gate)
data
closed (the README describes the training stages, but no training data is released or itemized)
code
partial (inference code and a LoRA fine-tuning script; the pretraining code is not published)
license
MIT (the GLM-Image weights)+Apache-2.0 (the repository code, and the X-Omni vision tokenizer and encoder weights the model incorporates)

GLM-Image's weights are MIT-licensed and the code is Apache 2.0, so the model can be used and modified freely. Only inference and LoRA fine-tuning code is published, not the pretraining pipeline, and the training data is not released.

Adoption

2 high confidence
2.0

Adoption is measured as Hugging Face downloads of the single GLM-Image checkpoint. Use through Z.ai's API is not counted.

Capability

2 medium confidence
2.0

GLM-Image places in the lower half of the public image arena, below FLUX.2 [dev]. Its strengths, text rendering and knowledge-heavy layouts, are a narrower slice than the arena's general prompts.

Verified 2026-11-12