Zhipu releases GLM-5 technical report with all technical details fully disclosed.

The Intelligent Chart states that GLM-5 is a next-generation foundational model designed to shift programming paradigms from "VibeCoding" to "AgenticEngineering". Building upon the intelligence, reasoning, and programming abilities of the previous model GLM-4.5, GLM-5 uses sparse attention to significantly reduce inference costs while maintaining long-range context capabilities intact. In order to align the model better with various tasks, a novel asynchronous reinforcement learning infrastructure has been constructed, decoupling the generation process from the training process and greatly improving post-training iteration efficiency. Furthermore, a new asynchronous Agent reinforcement learning algorithm has been proposed to further enhance the effectiveness of reinforcement learning, allowing the model to learn more effectively from complex, long-term interactions. Based on these innovations, GLM-5 has achieved state-of-the-art performance in mainstream open benchmark tests. Most importantly, GLM-5 demonstrates unprecedented capabilities in real-world programming tasks, surpassing all previous open-source baselines in handling end-to-end software engineering challenges.