GLM-5

GLM-5 is the fifth-generation large language model from Zhipu AI (Z.ai), featuring 745 billion parameters in a Mixture of Experts architecture with 44 billion active parameters per inference. It achieves near-frontier performance on coding (SWE-bench 77.8), reasoning (AIME 92.7), and agentic tasks (BrowseComp 75.9), while being 7x cheaper than Claude Opus 4.5. Trained entirely on Huawei Ascend chips, GLM-5 is open-source under MIT license. Originally released anonymously as "Pony Alpha" on OpenRouter, it quickly became one of the most talked-about AI models of 2026.