Today, the Zhi Spectrum GLM team revealed the first engineering practice progress in the direction of recursive self-improvement, which is to let AI improve its own operating system by itself. Specifically, the Infra Agent, driven by GLM-5.3, built and debugged inference services from scratch on a cluster of over 100,000 domestic chips, and increased the overall processing speed to 3 times the original in less than two weeks. The hardware utilization efficiency and single token cost are close to the same level as mainstream Nvidia GPUs. This system has already been put into actual use. GLM-5.3-Flash was launched on OpenCode and OpenRouter using the anonymous model OX-Alpha, and the number of token calls exceeded 62 trillion in 6 days. This is the first time that a major domestic model manufacturer has publicly used “AI self-improvement” in the production environment, which means that AI can no longer only write code, but also participate in a closed loop of engineering for the entire inference system.

Today, the Zhi Spectrum GLM team revealed the first engineering practice progress in the direction of recursive self-improvement, which is to let AI improve its own operating system by itself. Specifically, the Infra Agent, driven by GLM-5.3, built and debugged inference services from scratch on a cluster of over 100,000 domestic chips, and increased the overall processing speed to 3 times the original in less than two weeks. The hardware utilization efficiency and single token cost are close to the same level as mainstream Nvidia GPUs. This system has already been put into actual use. GLM-5.3-Flash was launched on OpenCode and OpenRouter using the anonymous model OX-Alpha, and the number of token calls exceeded 62 trillion in 6 days. This is the first time that a major domestic model manufacturer has publicly used “AI self-improvement” in the production environment, which means that AI can no longer only write code, but also participate in a closed loop of engineering for the entire inference system.