KNOWLEDGE ATLAS released the GLM-5 technical report, with all technical details disclosed

KNOWLEDGE ATLAS states that GLM-5 is a next-generation foundational model aimed at shifting the programming paradigm from "VibeCoding" to "Agentic Engineering." Building on the Agentic, Reasoning and Coding (ARC) capabilities of its predecessor GLM-4.5, GLM-5 employs DeepSeek Sparse Attention (DSA) to significantly reduce reasoning costs while maintaining long-context capabilities without loss. To better align the model with various tasks, a new asynchronous reinforcement learning (RL) infrastructure has been constructed, decoupling the generation process from the training process, thereby greatly enhancing post-training iteration efficiency. Additionally, a brand-new asynchronous Agent reinforcement learning algorithm has been proposed, further improving the effectiveness of reinforcement learning, allowing the model to learn more effectively from complex, long-range interactions. Based on these innovations, GLM-5 has achieved SOTA performance in mainstream open benchmark tests. Most importantly, GLM-5 has demonstrated unprecedented capabilities in real-world programming tasks, surpassing all previous open-source baselines in handling end-to-end software engineering challenges