Zhipu AI has launched GLM-5, a fifth-generation large language model with 745B parameters using Mixture of Experts (MoE) technology. With only 44B active parameters but the power to process large-scale data through a 200K token context window, GLM-5 represents a significant advancement in AI capabilities.
Training with Chinese Technology
What makes GLM-5 particularly interesting is its training on Huawei Ascend chips with the MindSpore framework, demonstrating the company's effort to create technological independence without relying on American hardware. This decision aligns with global trends seeking alternatives beyond NVIDIA GPUs.
Key Capabilities of GLM-5
GLM-5 excels in several areas crucial for modern AI development: Agent capability: Efficiently functions as an AI agent Code generation: Writes accurate and complex code Complex reasoning: Solves complex problems and provides sophisticated reasoning
Funding for Development
Zhipu AI conducted an IPO in January 2026, raising 43.5 million HKD specifically for GLM-5 development. This substantial investment reflects investor confidence in the potential of Chinese AI models that are growing in independence and efficiency. GLM-5 isn't just competing in the general language model market; it's showcasing the potential of Chinese AI technology developing in an independent and efficient direction. Reference: https://glm-5.org/zh/