GLM 5

GLM-5 is a large language model developed by zai-org, representing a significant advancement in AI, particularly for complex systems engineering and long-horizon agentic tasks. It scales substantially from its predecessor, GLM-4.5, increasing from 355 billion parameters (32B active) to 744 billion parameters (40B active), and expanding its pre-training data from 23 trillion to 28.5 trillion tokens. The model integrates DeepSeek Sparse Attention (DSA) to reduce deployment costs while maintaining long-context capacity. Furthermore, GLM-5 incorporates `slime`, a novel asynchronous Reinforcement Learning (RL) infrastructure. This infrastructure is designed to improve RL training throughput and efficiency, enabling more fine-grained post-training iterations. These advancements in both pre-training and post-training contribute to GLM-5's enhanced performance. GLM-5 delivers significant improvements compared to GLM-4.7 across a wide range of academic benchmarks. It aims to achieve best-in-class performance among open-source models globally for reasoning, coding, and agentic tasks, closing the performance gap with frontier models.