GLM-Image
16B open image model with best-in-class text rendering, trained on Huawei chips.
About
GLM-Image is Zhipu's open-weights image generator with two claims to fame. Architecturally it is a 16B two-stage system: a 9B autoregressive transformer that lays out the image, then a 7B diffusion decoder (built on CogView4) that fills in detail, with a Glyph-ByT5 encoder that models character shapes — which is why its text-in-image rendering beats open and proprietary rivals in third-party checks. The other story is the hardware: trained end-to-end on Huawei Ascend 910B chips with CANN and MindSpore, no NVIDIA anywhere, a milestone for China's decoupled AI stack. Weights are MIT-licensed and the API runs $0.015 per image. Caveats: training data is undisclosed, and Western developers should weigh ecosystem maturity (tooling, community) against the license generosity.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Third-party (DeepLearning.AI) reported: outperforms open and proprietary competitors on text rendering
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Zhipu open-weights image generator: 16B two-stage (9B autoregressive + 7B diffusion decoder), MIT license.; Trained end-to-end on Huawei Ascend 910B chips (CANN/MindSpore) — no NVIDIA; notable infrastructure story.; Third-party (DeepLearning.AI) reported: beats open and proprietary rivals on text rendering; API $0.015/image.; Limitations: training data undisclosed; ecosystem maturity vs Western alternatives..