Zhipu GLM Reveals AI Self-Improvement Breakthrough, Infra Agent Optimizes Its Own Inference System

Zhipu's GLM team has revealed the first engineering practice in recursive self-improvement, where an Infra Agent powered by GLM-5.3 designed, debugged, and optimized the inference infrastructure for GLM-5.3-Flash, enabling the model to improve its own operating system.

The Infra Agent built a production-grade inference service from scratch on a cluster of over 100,000 domestic chips, boosting end-to-end throughput to 3 times the initial baseline in under two weeks. Hardware efficiency and per-token cost reached levels comparable to mainstream NVIDIA GPUs.

The system is already in real use. GLM-5.3-Flash launched anonymously as Ox-Alpha on OpenCode and OpenRouter, surpassing 62 trillion tokens in 6 days. This is the first time a Chinese large model company has publicly applied AI self-improvement in production, meaning AI can now participate in the full engineering loop of inference systems.