Zhipu, in collaboration with Huawei, open-sourced GLM-Image, the first state-of-the-art (SOTA) multimodal model trained on a domestically produced chip.

Zhipu AI and Huawei jointly open-sourced their next-generation image generation model, GLM-Image, which is the first state-of-the-art (SOTA) multimodal model to complete the entire training process on the domestically developed Ascend Atlas 800T A2 chip. The model adopts an innovative "autoregressive + diffusion decoder" hybrid architecture and has achieved the best results among open-source models in the complex visual text generation (CVTG-2K) and long text rendering (LongText-Bench) leaderboards, especially excelling in Chinese character generation tasks.