Zhipu AI Releases 744B Parameter Open-Source Model
Chinese firm Zhipu AI has released GLM-5, a frontier open-source model with 744 billion total parameters and 40 billion active parameters via a Mixture-of-Experts architecture. The model was pre-trained on 28.5 trillion tokens using Huawei Ascend NPUs. Early reviews suggest it narrows the performance gap with Anthropic’s Claude Opus 4.5, particularly in coding and agentic tasks.
- Zhipu AI was spun out of Tsinghua University in 2019 and is backed by Chinese tech giants like Tencent and Alibaba, as well as multiple state-backed investment funds. - The model's training on Huawei Ascend NPUs is a strategic move to build a domestic AI stack, mitigating the impact of U.S. export restrictions on high-performance NVIDIA GPUs. - GLM-5
