Chinese AI firm Zhupu has partnered with Huawei to open-source a next-generation image generation model, GLM-Image, marking the first state-of-the-art multimodal model fully trained on domestic chips.

The model was trained using Huawei’s Ascend Atlas 800T A2 hardware and the MindSpore AI framework, covering the entire process from data preparation to model training. GLM-Image combines image generation with language model capabilities, allowing developers to generate images via an API for just 0.1 yuan (about 1.5 cents) per image, with a speed-optimized version scheduled for release soon.