GLM-4-32B-0414

GLM-4-32B-0414 is a 32-billion-parameter open-weight foundation model pre-trained on 15 trillion tokens of high-quality data, including synthetic reasoning material. Released under an MIT license, it delivers coding and tool-use performance competitive with much larger models, particularly in software engineering, web design, and structured content generation. The model targets code generation with function calling, artifact creation for code and SVG graphics, and search-based question answering with integrated retrieval. Its 32K context window supports long documents and extended development sessions, making it well suited for local deployment and fine-tuning. Specialized variants include GLM-Z1-32B-0414 for deep mathematical and logical reasoning and GLM-Z1-Rumination-32B-0414 for longer open-ended problems with integrated search. The family is available in BF16 and quantized formats across common inference frameworks.