Huawei processors were used to train Z AI's open image model

GLM-picture, an open-source picture generator recently released by Chinese AI firm Zhipu AI, is praised as the first significant model trained solely on Huawei hardware. Despite good benchmark scores, early testing has not been encouraging.

The specifics:

With no reliance on US semiconductors, the 16B-parameter model was created utilizing Huawei's Ascend chips and software.

Early user tests have not supported Z AI's claims that GLM-Image outperforms Nano Banana Pro on accuracy benchmarks and excels at text-heavy images.

In terms of total image quality, the model lags below leading alternatives like Nano Banana Pro and Seedream; however, it is made available as open-source software under a permissive license.

With a potent GLM-4.7 release in December (as well as a recent HK IPO), Zhipu established its AI presence. Now, it offers a capable, open image model that was trained completely without Nvidia. It shows that China's AI industry isn't waiting for the chip war to conclude, even though it might not defeat close competitors.