Search
On January 14, 2026, Zhipu, in a strategic partnership with Huawei, announced the open-sourcing of its cutting-edge image generation model, GLM-Image. Developed on the Ascend Atlas 800T A2 platform and utilizing the MindSpore AI framework, this model has successfully navigated the entire development journey—from data collection to training—on domestically-produced chips. This achievement marks GLM-Image as the first SOTA multimodal model to be fully trained on such chips.
Employing a hybrid architecture that combines 'autoregressive + diffusion decoder' techniques, GLM-Image seamlessly integrates image generation capabilities with language models. This innovation underscores Zhipu's pioneering efforts in exploring the new-generation 'cognitive generation' technological paradigm.
-
C114 Communication Network -
Communication Home
