Recently, in response to media reports on the training cost of Ant's Bailing big model, Ant responded by stating that Ant continues to optimize different chips to reduce AI application costs, and has made some progress. It will gradually share through open source. Previously reported, Ant Group successfully used domestic chips including Alibaba and Huawei, combined with a hybrid expert (MoE) machine learning method, to train artificial intelligence (AI) models, reducing costs by about 20%. The test results show that this technology can rival the performance of the Nvidia H800 chip.