[GLM Discloses China's First RSI Engineering Practice] According to reports, today the GLM team unveiled the first engineering practice in the field of Recursive Self-Improvement (RSI), driven by GLM-5.3. The Infra Agent, powered by GLM-5.3, completed the design, debugging, and optimization of the GLM-5.3-Flash inference infrastructure, achieving reverse improvement of the model's own operating system. This marks the first publicly disclosed RSI case implemented in a production environment by a domestic large model company. According to the official blog, the Infra Agent built a production-grade inference service from scratch on a cluster of over 100,000 domestically produced chips, boosting end-to-end throughput to three times the initial baseline in less than two weeks. Hardware utilization efficiency and per-token cost reached levels comparable to mainstream NVIDIA GPUs, while supporting a 1M context window and multimodal requests. The system has already been tested under real-world traffic, with GLM-5.3-Flash deployed as the anonymous model Ox-Alpha on OpenCode and OpenRouter, achieving over 62 trillion token calls within six days. The technical breakthrough of this practice lies in the fact that the Agent is no longer limited to generating code but can now execute a complete engineering feedback loop around the inference system: autonomously analyzing performance bottlenecks, identifying precision deficiencies, modifying underlying code, conducting layered testing, and iterating continuously based on experimental feedback. The GLM team stated that the system has not yet reached the stage of fully autonomous design and training of next-generation models, but the early form of RSI has already emerged—namely, the model builds the inference system, and the system, in turn, supports the model's operation. (21st Century Business Herald)