China's Z.ai Releases GLM-5 Frontier Model
Chinese AI company Z.ai has shipped GLM-5, the country's first public frontier model from a listed company. The model features 600 billion parameters and a 128k context window, positioning it as a direct competitor to Western large language models. The release marks a significant milestone in China's pursuit of AI self-sufficiency and is expected to accelerate competition in sovereign AI infrastructure.
- The model was reportedly trained entirely on Huawei Ascend chips, a significant milestone in China's "AI self-reliance" strategy aimed at creating a domestic hardware and software ecosystem independent of US technology and export controls. - GLM-5's architecture is explicitly designed for "agentic engineering," moving beyond simple code generation to support autonomous, long-horizon tasks; this is enabled by a 744B parameter Mixture-of-Experts (MoE) structure and a novel asynchronous reinforcement learning infrastructure called "slime" to improve multi-step reasoning. - Zhipu AI plans to release GLM-5 as an open-source model under a permissive MIT license, with weights already available on Hugging Face, while simultaneously raising the price of its GLM Coding Plan by 30% due to surging demand, indicating a dual strategy of community-driven development and commercial monetization. - The release comes as China implements its AI Safety Governance Framework 2.0, which uses a risk-based approach and requires lifecycle management for AI systems, creating a distinct compliance and regulatory landscape for companies deploying models in the region. - While a recent survey found 52% of Chinese CEOs report that AI has boosted revenues (a figure far exceeding the global average of 29%), other studies show only 9% of firms realize significant value from AI, highlighting a gap between enterprise adoption and scalable, profitable implementation. [38
