A Chinese AI lab has just released a model that can optimize code over 600 iterations without human intervention. GLM-5.1 increases the query speed of a vector database from 3,500 queries per second to 21,500 queries per second. It built a Linux desktop in a browser in just 8 hours, adding a file browser, terminal, text editor, and games—all integrated into a unified user interface. So, can it beat GPT?