[GLM discloses the first RSI practice of domestic large models: AI builds its own inference system on a domestic cluster of 100,000 cards] "Science and Technology Innovation Board Daily" reported on the 17th that today, the GLM team disclosed its first engineering practice in the direction of recursive self-improvement (RSI): Infra driven by GLM-5.3 Agent completed the design, debugging and optimization of GLM-5.3-Flash inference infrastructure, and realized the reverse improvement of the model to its own running system. This is the first time that a large domestic model manufacturer has disclosed an RSI case that has entered the production environment. According to the official blog, Infra Agent completed the construction of a production-level inference service from scratch on a cluster of more than 100,000 domestic chips. In less than two weeks, the end-to-end throughput increased to 3 times the initial baseline. The hardware utilization efficiency and single token cost reached the same level as mainstream NVIDIA GPUs. (Kechuangban Daily reporter Li Mingming)