GLM-4.5: Open Agentic, Reasoning, and Coding Foundation Models

What this episode covers

The source introduces GLM-4.5, a new open-source Mixture-of-Experts (MoE) large language model, along with a compact version, GLM-4.5-Air. Developed by Zhipu AI and Tsinghua University, these models are designed for agentic, reasoning, and coding (ARC) tasks, exhibiting strong performance in these areas despite having fewer parameters than many competitors. The paper details their multi-stage training process, which includes extensive pre-training and post-training with expert model iteration and reinforcement learning, enabling hybrid reasoning modes for both direct and deliberative responses. Furthermore, the source presents comprehensive evaluation results across numerous benchmarks, showcasing GLM-4.5's capabilities in general chat, logical reasoning, and complex translation, alongside insights into their RL infrastructure and data synthesis methods.

PodParley-generated summary based on available episode metadata and transcript content.