According to CoinMeta, a Saudi company recently released a model, initially calling it the “100% Saudi Model.” After facing doubts, the publisher clarified that the model was developed based on a model from a Chinese company under its umbrella. Both models adopt a hybrid expert architecture, with a total of 428 billion parameters and 23 billion active parameters. It is stated that the model underwent additional training using over one trillion Arabic words, with a focus on enhancing Arabic language skills, local culture understanding, intelligent agency capabilities, tool utilization, and computer operation abilities. The model achieved an average score of 89.37% in seven Arabic benchmark tests, and there are plans to make the model’s weights publicly available and deploy it on Saudi servers. Experts believe that this collaboration indicates that Chinese models are moving beyond providing basic services and products to becoming a technical foundation for other countries to build their own sovereign AI systems.